It is simple to understand why. The sharing economy was predicated on monetizing slack time in assets—we’ve all heard the riff that cars are parked 95% of the time. Uber worked because cars were valuable enough to justify driving, but most other sharing economy companies failed because they required too much effort.

Waymo, and self-driving cars, are going to change this and transform our economy into ones where our consumer goods become true assets generating passive income.

The Economics of Sharing Cars

The sharing economy bifurcated in two ways. The companies that worked generated cash streams that were high enough to be someone’s actual job because they involved the two most expensive assets people own, their homes and their cars, in transactions that happened frequently enough to justify the effort. On the other end were assets that were not valuable enough or frequently used enough to generate cash streams high enough to replace your job. Those assets need to be basically passive to justify the investment, which means the transaction costs need to be low. That last part is where the sharing economy failed: it turns out most assets aren’t that valuable to monetize, and the fixed transaction costs made monetizing them anything but passive.

What we got instead was mostly the gig economy. Companies that made it easier to monetize your time, not your stuff, were what actually took off. Rent the Runway has never posted a profitable fiscal year, but Fiverr went public; E Umbrella went bankrupt due to theft, but Taskrabbit was bought by Ikea for a billion dollars. The sharing economy, in this way, did work—it just required us to still be part of the equation. The big kahuna here, of course, was Uber. It allowed people to monetize their cars, but it still required them to have the spare time to drive. In fact, for a long time, this was part of Uber’s marketing to drivers.

Many fear that Waymo will lead to a loss of jobs. While it clearly will in the driving world, it will actually lead to a net increase of jobs, and more importantly, give those drivers passive income.

Waymo is obviously a new paradigm. Waymo’s cars have already driven over 96 million miles this year and are already over 90% safer than humans, according to their insurance provider Swiss Re. The cost curves have hardly even begun to come down and Waymo’s hardware costs are already just 30 cents per mile, according to its CEO, compared to $2-3 for Uber and Lyft. Add in extraordinary utilization rates—according to Uber, Waymos are in the 99th percentile compared to human drivers—and you have a moneymaking machine on your hands. Waymo has a mere 800 vehicles in operation in the Bay Area compared to tens of thousands of rideshare drivers and is already bigger than Lyft. But Waymo isn’t going after the rideshare market, of which it currently controls about 25%, but total miles driven, which is about 5 times bigger in San Francisco (and lower elsewhere in the Bay). In other words, Waymo is just at 5% market share of total vehicle miles. That will require quite a few cars! No wonder Waymo is exploring personal vehicle systems, starting with Toyota. Of course they are—this will drive economies of scale for sensor components like LiDAR and help satisfy consumer demand.

This will also obviously lead to the ability for people to rent out their cars on the Waymo network in exchange for revenue share net of operating costs, namely cleaning, maintenance, and refueling. Waymo will welcome this. The reason asset light is such an appealing business model is that it is also liability light. For cars especially, auto loans are generally lower for personal use. Consumers will welcome this too. Right now, the average Waymo generates about $200 in revenue per day, or $6000 per month. The average consumer auto payment is just $748.1

Even though the cost of a Waymo will go down, and even though there will be maintenance costs, the math still pencils out for consumers. Say there is a 50% gross profit—that’s still $3000 per month, far exceeding the monthly car payment. Yes, prices will go down from here, but utilization will go up, resulting in a wash if not a modest increase. That means an extra $1000-2000 of spending money a month, not to mention a paid-off car for when you want to use it.

Uber drivers will also love this—the vast majority of them are part-time, meaning that Uber is a part-time second job. The only thing better than making money from a second job is making money passively.

Waymo will also result in new jobs for the 9% of ex-Uber drivers who are currently full time (on top of the new passive income stream!). People won’t rent their cars out unless it comes back clean, which will create enormous demand for logistics jobs and cleaning jobs. Cleaners make $20-30 an hour, while the average Uber driver in California makes $21 an hour with more risk. These cleaning jobs will certainly command a premium because the required standards will be very high. Furthermore, because Waymos will be a replacement for all cars, not just rideshares, the logistics demand will be significantly higher than if Waymo merely replaced Uber.

Not Just Driving

That’s not the end of the passive income story, however.

Waymo will be so transformative that it will also lead to the creation of new, monetizable assets. The most obvious one is garage conversions. With self-driving cars shuttling around, it will not make as much sense to have parking in offices and homes. Today, it is standard for building codes to require parking. That is going to go away. It’s already starting to. When that happens, you’re going to see a wave of garage conversions. Who wouldn’t want all that extra space? A garage takes up 12-15% of a home’s square footage, and 55% of people already do something else in their garage like exercise or work. Today, a garage is needed to support your home value, but it won’t when you and your guests don’t need to park. That converted space will also be monetizable. How many of those new rooms will be Airbnbs? How many converted office parking spaces will have popups or logistics centers?

This won’t just apply to Waymo. The general idea of our devices getting smart is going to apply to more of our personal possessions. Expensive personal robots are coming into our homes. Ultimately, they will suffer from the same utilization problem that cars do. They’ll be too useful not to have, but lay dormant 90% of the time (there’s only so much cleaning to do). If your robot is as expensive as your car,2 you’re going to want to rent it out. Yet more passive income.

Even here, Waymo itself rears its lovely head. If you want to rent out your Sunday Robotics cleaning robot during its downtime, without cost-effective logistics, you won’t be able to make any money. With Waymo, the cost of shuttling your cleaning robot drops 80-90% and can be done on-demand with little to no input from you. This not only drives more income and utilization for those rented out Waymos—it creates a business model for Sunday Robotics and allows its customers to monetize more of their assets.

The Golden Future

This passive income is coming just in time, too. It is a sort of market-enabled UBI just as a giant productivity boost is coming that is going to reduce the amount we work significantly.

Like it or not, the four day work week is already here. According to Ari Emmanuel, the top day for booking hotels is now Thursday, not Friday. In a few years, intelligence from LLMs to AVs will bring us to a three-day workweek. This is what happened during the Industrial Revolution too. Productivity increases led to a steady decrease of work by 4 hours per week each decade, going from an average of 69 hours per week in 1830 to 42 hours by 1930. By the time the 40-hour work week became law in 1938, “Fordism” was already the norm.3

The passive income society will help power this. It will also help people get started or get by during hard times. While we often focus on get-by jobs, like driving a cab or serving coffee, people also make money by renting out their stuff. The comedian Andrew Schulz, for example, sublet his apartment and slept in his closet while he was getting started. This is what will replace the low barrier to entry “getting by” jobs: low cost of entry “getting by” assets.

The new consensus is that political opposition to AI is coming from the AI industry not doing enough to explain the benefits of their vision. This vision is clear: AI will make us safer, work less, and turn the average person into a rentier.

The passive income society can’t come soon enough!

These days, it seems like new technologies face an unprecedented onslaught. Political opponents, astroturfing, histrionic media stories, and more. It can make you feel like these people are stopping progress. We see it for many new technologies, like nuclear power, supersonic flight, and autonomous vehicles, all slowed down by regulation caused by fear mongering. We think that if only we could make people see the light things would move faster.

Then I learned about artificial ice.

For most of history, ice was harvested in the winter and stored throughout the year. Amazingly, we got the Model T before we stopped using natural ice.

And it turned out that artificial ice faced its own resistance. There are some interesting lessons in there in how progress happens, but it leaves a brooding mystery for us to solve to unlock progress in our own era.

The Importance of Ice

Since ancient times, humans have used ice to store and preserve food. There are Bronze Age Akkadian records of ice houses that would store ice collected during the winter for summer drinks and ancient Egyptian clay pots from 500 BC that are believed to have stored ice. Yet without the ability to create artificial cooling, the only way to get ice was to harvest it, which of course made it very expensive. Whether Aztec, Akkadian, or Austrian, ice was used largely by the rich. If the winter was not cold enough, an entire region could face an “ice famine.”

However, the rise of American population centers in relatively hot climates created enough demand to foster a robust ice trade starting in the early 19th century, pioneered by “Ice King” Frederick Tudor, peaking at nearly $500 million.1 Chilled drinks, once the province of kings, spread to the masses. Previously-infeasible foods like ice cream became widely accessible. It became possible to transport produce in such quantities that special rail cars were developed. Medicine cold chains were literally life saving.

But of course, when society becomes so dependent on ice, depending on natural cycles becomes less tolerable. Enter artificial ice.

Technologically, artificial ice is an invention centuries in the making that solved this serious societal problem. There have been many iterations of technologies, from a Spanish physician who used salt to lower water temperatures to more modern compression refrigerant systems. The key breakthrough was the discovery of the vapor-compression cycle by William Cullen, which is how refrigeration still works today. Decades later, John Gorrie realized this could be used to make ice. Various improved systems were proposed, varying in which refrigerants they used, the design of the components of the system, and so on. The first practical implementation was built in Australia, where it was a commercial success due to the impracticality of shipping ice great distances from Britain. Although artificial ice was initially very expensive, costs came down over several decades,2 eventually resulting in the cessation of ice harvesting. Most people today wouldn’t even realize that for most of human history ice was only found in nature.

But along the way, artificial ice faced a punishing scare campaign. What is so notable is how similar to what we see in the modern day.

Scandal and Slander

Several states have decided to ban artificial meat. Those who wish to do so scare and recruit sympathetic farmers to talk about how how lab-grown meat is “unnatural” and scare consumers into thinking that meat is unclean, all in support of legislation banning such products. That kind of coordinated campaign is exactly what happened with artificial ice.

In his book How We Got to Now, Steven Johnson explains how the ice industry—which coined the term “artificial ice”—spread rumors about the new invention, even claiming it could carry disease. It didn’t help that the Ferdinand Carree ammonia refrigerators were the ones that first took off. Ammonia-produced ice would leave a white residue when it melted. Ice plants could explode or have ammonia leaks, leading to calls for regulation. Natural ice manufacturers ran vicious ads, like this one from Lindon Ice that says “Science Fails And We Must All Bow Our Heads To Nature.”

The newspapers were part of the effort to keep natural ice alive. After all, an 1880 New Orleans newspaper article noted that there remained a “natural prejudice” in favor of natural ice. Even in 1911, nearly 50 years after artificial ice appeared, a newspaper in Scranton praised the virtues of artificial ice. A Philadelphia newspaper published an elegy about a sympathetic family’s ice harvesting business. Fears were spread about the fitness of artificial ice for human consumption, as memorialized in an 1889 academic article from the Texas Medical Journal; never mind that natural ice had greater health concerns, like water runoff pollution and typhoid.

The assault in the press was so severe that Gorrie could not attract investment and died penniless, perhaps delaying progress in refrigeration by decades.

The natural ice industry coordinated this decades-long campaign alongside corrupt politicians. Charles Morse formed a monopoly called American Ice Trust with the support of New York City’s first mayor, Robert A. Van Wyck, increasing prices in the Ice Trust Scandal that helped lead to the downfall of Tammany Hall.

They formed the Natural Ice Association of America, which was a combination of what today we would call a PAC and an astroturf operation. According to the Natural Ice Association’s first meeting minutes, it was primarily focused on creating a false debate around cleanliness. For example, it hired W. T. Sedgwick, a professor at MIT, to produce poppycock like the claim that water tends to purify itself in freezing, and H.W. Mill, the director of epidemiology for the Minnesota State Board of Health, to argue that the apparent association between natural ice and typhoid was likely caused by other contaminants, like dirty hands. This is not unlike various energy groups that fund anti-green energy groups and get politicians to say their stuff is “good for humanity,” or anti-tech groups that scare people into vastly exaggerated concerns about the water usage of data centers.

All We Can Learn is a Question

But here’s the interesting thing: the scare tactics didn’t work.

Society had come to depend more on ice as it urbanized and industrialized. From 1880 to 1914, American ice consumption tripled. That made ice famines less tolerable. An 1898 British ice famine, for example, was so severe that it led for a national call to develop ice plants.3 Thanks to exposure, public opinion turned increasingly positive towards this always-exciting technology. In 1914, artificial ice finally overtook natural ice. More ice was produced than harvested. Society never looked back.

Zooming out, this pattern is not unique to artificial ice. Looking back, in fact, it seems like most technologies have faced these types of headwinds. No less than Frederick Douglass helped lead campaigns against vaccines. Thomas Edison used to scare people away from AC currents by using them to electrocute animals and newspapers published cartoons about the horrors of technologies like electricity. Even the Luddites weren’t unique; the word “sabotage” comes from French Luddites who threw sabot shoes into industrial machines. Decelerationists, in other words, will always be with us.

Despite all the fear-mongering, we produce goods in factories, we vaccinate our kids, we don’t use DC for all our electricity needs, and yes, we drink Coca-Cola with artificial ice.

What is clear from the history of technology is that everything that impedes progress now—astroturf fear mongering, political corruption, legacy media nostalgic sob-stories, hobbling anti-progress regulation—attempted to impede progress in the past too. These tactics were used because they appeal to timeless fears and weaknesses in human society and, perhaps, even the nature of our souls. These techniques are therefore timeless and we should not be surprised at all that they are emerging today in response to new technologies.

What is remarkable is that these various scare campaigns, all of which corresponded to calls for boycotts and bans, failed. Today, however, activists use the same tactics to great success, resulting in legislation against lab-grown meat, nuclear power, and much more. The question is therefore not how to stop spasms of technophobia but why the same attacks are more successful now than they were a century ago.

The answer does not seem to be that we are too far from industrialization; these techniques did not seem to work that well until recently. Children watched cartoons as recently as the 1950s like “A is for Atom” extolling the marvels of nuclear physics, even after the despairs of the “industrial wars” we now know as the World Wars. It is impossible to imagine that today, but it was not so long ago a woman could remark on how “wonderful how DuPont is improving on nature” with their chemicals at the 1936 World’s Fair. “What happened in 1971?”, indeed.

It seems like fear campaigns failed for the simple reason that society had a different attitude towards progress. But that does not answer why they felt that way.

Perhaps it has to do with the immediacy of the fears of the past. A century ago, people were worried about whether they would get sick from typhoid and die. Today we worry much more about long-term effects that are uncertain and require decades to measure. As Virginia Postrel put it, “To our 19th-century ancestors ‘nature’ was a source of peril rather than purity.” The problem is that positive attitudes towards technologies continued far past the point of subsistence industrialization.

Today’s fear campaigns are different. They are more amorphous. The result is a strange situation where only a rare injectable meme actually stops progress of the thing itself (say, nuclear power), but they do almost all contribute to a general environment that conspires to limit progress in general.

When Teslas were new, there were stories about electric cars catching fire even though they are 11 times less likely to catch fire than a gas car. On the one hand, electric vehicles live on. On the other hand, people still live with this fear, and it has slowed adoption in the United States. This is not the only example. No matter how many times vaccine myths are debunked, there is a subset that stubbornly refuses to get vaccinated because they have sincerely come to believe that Bill Gates wants to give their kids autism and turn them into 5G towers. Notably, even though the original Wakefield MMR-autism article has been disproven, many people still believe that claim and have even gone on to other unrelated vaccine related fears, like dosing schedules. Even for technologies where exposure decreases fear, like autonomous vehicles, attempts to ban them continue. The deep question is how to make people into what Peter Thiel calls definite optimists.

Perhaps we have simply adopted another value system. Way back when, people made much more direct trade-offs: am I better off, regardless of whether the cause is natural or invented? Today, we seem to value purity and assume that natural is better. This could be true, but it would still be circular. We would still not know why we came to adopt this new value system at the expense of the progress that got us there.

The original article calling for progress studies by Patrick Collison and Tyler Cowen called for a better understanding of the causes of progress. While it is true that we do not know why the Industrial Revolution happened when it did, the episode over artificial ice suggests, as Joel Mokyr argued in his now-Nobel Prize winning book A Culture of Growth, that the key factor is the attitude of people. Perhaps the most important question for progress studies is not one of studying institutions and the network structure of universities but rather solving the social psychology of fear of the future. What this vignette suggests is that the reason progress has stopped is not some arcana surrounding science funding but the simple answer that we wanted it to stop. And that is a most frightening prospect.

As of now, there is no satisfactory an answer to the question: why are we so much more fearful that anti-progress propaganda can work and so much more resistant to good news? Answer that and you might unlock a whole lot of progress.

Trust seems similar to brand, but it has a subtle difference. Brand is always an advantage for incumbents—it reduces acquisition cost, improves competitive positioning for risk-averse prospects, and so on. In low-trust markets, you may have powerful incumbents, but oftentimes they will be chosen for reasons of better-priced bundles or numerous features. But trust creates a market where there is an acceptable set of vendors, regardless of whether they offer the product, and companies outside the competitive set are not considered at all regardless of the offering. High-trust industries lead to a “nobody got fired for hiring IBM” situation on steroids. The key sign of a high-trust industry is a small number of vendors who offer a large number of products where your product is not even compared unfavorably but not even considered at all; once you start to get through the door, you will encounter the same names every time. Though this difference is subtle, it feels very different.

Subscribe now

The key factors for trust are:

1. How high stakes are mistakes? Is a mistake potentially catastrophic, or just part of doing business? Sometimes, there is a probabilisitic aspect, where normally something doesn’t matter, but a few times matters a lot.

2. How irreversible are mistakes? Even a very high-stakes error suddenly becomes less scary if you can undo it, but even small mistakes can be fatal if they are irreversible and numerous.

3. How verifiable are mistakes? Sometimes you can tell just by using a product how good it is; verifiability is much harder if the use case involves counterfactuals, high expertise, or long feedback cycles.

4. How accountable is the mistake-maker? A lawyer or doctor could go to jail. A cybersecurity loss is often insured.2 A scientist might be rewarded for conducting a failed, but hard, experiment.

Trust is ultimately a spectrum. On the absolute low-end are businesses that require essentially no trust to operate at all. On the absolute high-end are businesses that no one will use without validation and broad market acceptance. Most businesses fall somewhere in between or have different aspects of the business with different trust levels. It is branding meets network effects where the network is the visibility of the customer base.

Low trust example: CRM tools. While modern CRM software can be complex, it falls into the category of “CRUD database with a nice interface and business logic.” This is not to demean the quality of those products and the work required to build them—simply to point out that you can assess the quality of the product quickly by using it, and that the cost of a single failure (such as a lost record) is low because of its high volume of use. The decision to use Hubspot, Salesforce, or Zoho is not about whether you trust it but rather which segment you fall into.

Medium trust example: payment processors. Even though payment processing deals with the most fundamental part of business (accepting money) and thus requires some trust, it requires less trust than most other financial infrastructure because of the high frequency and low stakes of most transactions. Unless your business is failing, your payment processor will be accepting a credit card several times a week, if not several times a minute. There is a tight, closed loop, making the immediate value easier to verify and the blast radius of a mistake limited. Companies will generally want to use a well-known payment processor, but there will always be a market for a new player because there will always be some people willing to try something new to save a buck. The limiting factor is that payment processing is hard.3

High trust example: e-discovery tools for lawyers.4 E-discovery is a form of search where the great fear is that you would miss something. While you can verify the quality of the results (at great cost),5 you cannot by definition verify the quality of the counterfactual. For example, in a recent copyright lawsuit against Meta, their e-discovery tool failed to produce 18,000 relevant documents. This one mistake exposed the lawyers to a lot of liability, including sanctions and malpractice suits, and jeopardized their entire client relationship through bad publicity. This one failure more than nets out any time savings of even 99.9% accuracy.

When lots of trust is involved, buyers will certainly shop around, but they will typically only consider vendors from a small list of well-known players in an industry. For this reason, a portfolio of products is ultimately the winning strategy in trust-based industries. The reason is not distribution; it is that someone is always looking to buy something, but if they are only willing to consider a small number of vendors, they will reach out to those vendors to see if they have something at all. If there is a small universe of acceptable vendors, if you are the only one with a certain offering, you then win by default.

High-trust products are always mission-critical. They are currently in use and people will keep using them because, no matter how ugly or inefficient or buggy it is, that solution is working. The job is getting done. So “better, faster, cheaper” isn’t enough to get the buyer to move over because of the counterfactual risk. This is not to say that trusted vendors are perfect, but their error rate is perceived as accurate and their ability to withstand mistakes is viewed as believable.

Thus, trust-based buyers will only consider investing time in even trying vendors where they believe there is some chance of success (and low downside). The proof of your competence is being widely used, so the obvious question becomes, how do you get your first reference customers?

What you need is a platform shift. A platform shift creates the potential for a totally new experience that is either insanely better or introduces a new dimension of quality altogether. Incumbents in trust-based industries are usually slower because the competition is less. While you will likely not eliminate them, if you act quickly, you will become one of the trusted vendors.6 While platform shifts usually come from new technologies, they may also come from sudden changes like a dramatic, unexpected regulatory change.

The early sign of success is getting key logos who simply cannot get the default vendors to offer something to them. Like a doctor in residency, the customers become the startup’s tutor and lend their institution’s reputation to the next prospects. The next sign of success is them asking you to offer them more products. This is a sign that they trust you. Trusted vendors offer lots of services, and if they treat you that way, they believe you will be around for the long run. If they are telling other people in their space—and in high-trust industries, the community is usually small—that is proof that you’re really on to something.

Breaking into a high-trust industry is, to some extent, out of your control because you need a window that causes several customers to be open to you at once. While there will always be some Crossing the Chasm-type early adopters, being an innovator is generally a personality type, not a profile, so your likelihood of finding those people is vanishingly small without the platform shift. If you can, it is better to launch with several design partners out of nowhere, ideally in a coordinated fashion, to establish dominance.

The nice thing about breaking into a high-trust industry, however, is that once you’re in, you’re in. These windows tend to be short, so it will be difficult for other startups to compete with you unless they also broke through during the same window. Trust-based industries also tend towards consolidation, so if you make it through the window, expect to receive acquisition offers. Pay attention to everything your customers ask you for because you will likely offer it as a product in the future.

No one ever got fired for buying IBM, but if IBM isn’t even selling the new thing, you aren’t even competing with them, so high-trust customers will be willing to try you out. By the time IBM decides to compete with you, if you’ve successfully established yourself, you’ll at least be one of the viable vendors buyers will consider. Though it is easier for IBM to just buy you.

“The best lack all conviction, while the worst/Are full of passionate intensity.”

William Butler Yeats, The Second Coming

Abundance, by Ezra Klein and Derek Thompson, is the newest and most anticipated release of a series of books that have recently come out on the topic of improving progressive governance, like Dunkelman’s Why Nothing Works, Demsas’s On the Housing Crisis, and Appelbaum’s Stuck. Particularly given the authors’ fame, Abundance has triggered a significant debate, much more than the other books in this subgenre.

Abundance exists in the context of a conversation the authors helped start that focuses on maximizing outputs rather than inputs. It goes by several names, like “supply-side liberalism” or “the abundance agenda,”1 both of which are articles written by Klein and Thompson around the same time. To put it briefly, these ideas argues that progressivism focuses primarily on subsidizing demand for goods that may not be available to the poor or middle class yet is shockingly unconcerned with whether those goods are actually delivered—even worse, many progressive regulatory projects make it harder to actually achieve those goals through undersupply (hence the name) all while patting themselves on the back for providing funding to tackle these initiatives, regardless of those goals are achieved. This idea broke further into the mainstream with ’s essay on “checkism,” which is the idea that progressives evaluate projects by whether a check is written and not whether the goods are delivered, incentivizing less effective projects over time.

Yet, Abundance doesn’t merely recount this years-long dialog; it adds to it. Firstly, although these various ideas have been floating around, no one has yet put them into the form of a coherent agenda that connects them together in a clear way. You might summarize the Abundance perspective as: “if you promise something, you better deliver it.” But Abundance is also a deeper project that asks its readers to demand something different from government. The typical progressive goal is one of negation: anti-racism, de-growth, no-pollution, and so on. They are positions of scarcity with policy solutions that emphasize redistribution. This leads to what Klein calls “everything bagel liberalism,” a phenomenon whereby progressives are unable to choose between priorities, ultimately stymying progress in all of them.

Abundance wants to redefine the goal of government from stopping problems—again, a scarcity mindset—to pursuing abundance. This might seem like a distinction without a difference, but consider someone who believes in “managed retreat,” or the idea that we should simply abandon certain communities for environmental reasons. Managed retreat proponents try to use the government to stop development in places where they believe humans should not live. Abundance would not argue about the most effective way to achieve managed retreat—rather, it disagrees with that goal entirely and argues that it is not actually the progressive position.

The book focuses primarily on the obstacles to built infrastructure, like housing, transportation, and green energy. It divides the book into five chapters, each of which is a core tenet of governance: grow, build, govern, invent, and deploy. Each chapter focuses on vignettes which illustrate success or failure in its category (for example, the “invent” chapter focuses mostly on medical innovations like mRNA vaccines). The purpose of these examples is to show how government policies may be intended to help achieve these goals but primarily get in the way by failing to ask what is needed to achieve these goals. So for healthcare, for example, they praise Obamacare for making it easier to get insurance yet criticize progressives for not caring about the fact that government must also incentivize the creation of new treatments for that health insurance to buy in the first place.

One also senses a subtle, resentment-tinged riposte of the commentariat class. When Zephyr Teachout criticizes Klein and Thompson for being confused about what they really want, and Thompson calls out antitrust theory by pointing out that it is a failed model of explaining the world, asking “what’s oligarchy doing for you?” Indeed, it fails to explain why Texas passed SB 15, a bill almost identical to California’s SB 9, almost entirely without discussion. More importantly, the Abundance perspective is, who cares about antitrust? What matters is the number of houses.

The volume is pretty slim. Although it promises a sweeping idea, it reads like a mix of technocratic, somewhat-narrow policy proposals. To some extent, it is because they choose to use vignettes to depict their ideas but don’t expand much beyond the vignettes themselves, leaving the bigger picture thinking as an exercise to the reader. But partly, I think, it comes from a deeper problem.

If this was all the book was, I would have high praise for it.2 I still think it can serve a useful purpose in decreasing the derangement that has gripped progressivism. Yet I suspect that it cannot finish the job because it fails to believe in its own idea enough.

Can You Avoid the Non-Progressive Bundle?

In some sense, the book is an attempt to answer the question “Why are so many Californians leaving for Texas?” The sense of frustration is palpable. California, my home (where Klein grew up and lived for several years), is the punching bag of Abundance as much as it might be on Fox News as the authors lament the movement to, of all places, Texas. If you were to read their book with no critical eye, what you think you are walking away with is an answer to this question. What you are actually walking away with is an attempted supply-side answer to the question “why can’t liberals enact their own priorities?”

So what’s the problem? These are not actually the same question. Abundance fears to countenance that terrible possibility that one cannot merely tweak zoning laws and create a liberal utopia, that in fact to make the whole thing work, both from an outcome and political economy perspective, one may have to adopt a much bigger bundle of policies and attitudes that would cut quite counter to their progressive beliefs. Abundance promises a philosophical shift that prioritizes outputs, but the primary content of the book is mostly a series of adjustments to a still fundamentally progressive project. But Texas is not just California with more rational zoning. The bundle of policies people are choosing is much, much broader.

If you look at polls of why people are leaving California, it is not just housing costs. The most important reason people leave is a broader spectrum of the cost of living. Rolled into the idea of cost of living is, implicitly, job opportunities (after all, what does a price mean without an income?) which effectively comes down to business climate. The other top reason cited below cost is crime, which is never addressed in the book.

Let’s take the issue of cost of living. Yes, housing is the most expensive part of any budget, but the cost of living also includes the cost of food, gas, electricity, insurance, and many other goods. It is not just whether the government can build trains. Here, California is abysmal. Californians pay very high costs for food and transportation. Despite producing so much solar power it cannot use it all,3 Californian utility rates are the second-highest in the country. Gas costs are the highest in the country. Food isn’t expensive because of zoning. Neither are utilities.

If you were to take Abundance at face value, you’d think that San Francisco would be flooded with housing if only it would loosen zoning. Yet, this hasn’t happened because there are other frictions in the market. To get abundance, you have to reduce the costs of all the bottlenecks, not just the ones you find politically pallatable, and Klein and Thompson do not seem willing to even consider whether that requires a greater reconsideration of their views. This is all important because the authors’ goals will be unattainable unless the vendors available in the economy are reasonably priced and free to get the job done quickly and well. For example, San Francisco recently passed a law that made the permitting process for office-to-housing conversions easy. So why aren’t they happening? The cost of building is too high to make it profitable. It wasn’t just enough to fix the zoning rules—San Francisco has a separate bundle of progressive policies that stop abundance from happening.4 Abundance is very concerned with all the paperwork the government has to go through to build infrastructure, but seems surprisingly incurious about the 420,434 regulations in California that primarily burden the private sector and add to the costs of everyday life. This, too, is part of abundance.

The problem is this: when people think of abundance, they don’t just think about infrastructure. Abundance, yes, means that you can easily find an affordable place to live and take a subway. But it also means that you can afford a car, that you can eat out at a restaurant, and that you can find a well-paying job, and this is also a place where Abundance seems disinterested in breaking from the progressive project. A Hoover Institution analysis found that government and publicly supported sectors grew employment by 316,000 while the regular private sector lost 154,000 jobs. That has nothing to do with zoning laws or building restrictions, but it is indicative of general government restrictions on private industry that results in fewer opportunities for regular people—and opportunity is part of abundance, too. California’s business formation is one of the lowest in the nation. Could it have something to do with the fact that California has the 5th highest state and local tax burden in the country? The states with high employment growth have lower regulatory and tax burdens, and also tend to have laws on the books that Klein and Thompson would find inconvenient, like “right to work” rules.

Generally, Klein and Thompson seem pathologically resistant to accountability, either for people or for ideas. They quote Heidi Marston, the former head of the LA Homeless Services Authority, as saying that if she only had less accountability, she would have been able to spend $1 billion a year more effectively. But is that really believable? Her views on homelessness focus on discredited ideas like housing first while she oversaw one of the worst explosions of homelessness in LA history,5 and her entire career has been spent in non-executive roles with no demonstrated background in any area relevant to homelessness. How does the “personnel is policy” group believe the answer is to have less accountability for members of the Nonprofit Industrial Complex6 given the results? Similarly, their explanation for why California’s High Speed Rail project’s insane route is an obscure Obama-era funding rule and overzealous environmental authority, but the real reason is political horse trading, some of which was nakedly corrupt. Why don’t they call for any accountability for these decision makers? And even if the technocratic explanations were correct, why does Abundance not call for, at a minimum, a firing?7

I praise the get-done attitude of Abundance, and wish I would have seen one more level of belief in their own idea. The authors seem to often take the easy route. For example, they are happy to criticize regulations that YIMBYs dislike, but not ones that are more broadly popular among progressives. You see them clamoring for zoning reductions, but they don’t call for an end to rent control, which also increases housing prices. They’re happy to point out how much of San Francisco is zoned for single family homes, but won’t note that 60% of San Francisco is subject to extremely strict rent control. They also don’t seem Abundance-brained to have ideas progressives wouldn’t care for. One common NIMBY critique of high-rises, for example, is traffic and parking. The Abundance answer to this is “deal with it,” but why not proffer the capital-a Abundance Answer of “build tons of parking structures and subways?” With NEPA and CEQA, Abundance seems primarily concerned with the easy cases, like how NEPA absurdly makes it easier to frack for gas than build solar panels. But what about the actual, social trade-off between production and polution; shouldn’t we at least get a framework for thinking about this? When the rubber hits the road, one wonders, would Klein and Thompson ever, even once, choose to cause the extinction of an endangered fish if it meant building a car plant?

Klein and Thompson are quite clear that they are tackling the topic of abundance from the left—in fact, on Bari Weiss’s Honestly podcast, Thompson still calls himself a “tax-and-spend liberal.” Fair enough. But does that mean that they lack conviction to go after their historical political allies, especially when, as Dean Ball put it, many governments now seem to be “in business for [themselves] rather than for [their] citizens?” While they bemoan spurious CEQA lawsuits by NIMBYs, approximately one-third of all CEQA lawsuits come from unions, the single largest group of CEQA abusers, which they use to “greenmail” developers in placing no-bid contracts with union members. This practice receives nary a mention in Abundance. In fact, Abundance doesn’t talk about any supply-side improvements relating to labor, probably because these would mostly impact progressive constituencies. You won’t hear anything about the environmental professor who invented fake endangered species to stop developments he didn’t like or any potential tweaks to the Endangered Species Act—you won’t hear at all about fixing academia’s culture of fraud at all. You’d never know that when Wisconsin fully decertified their teacher’s union it improved public school performance. Aren’t good schools part of abundance? Evidently not, or at least not important enough to make the book.8

The most common phrase in the book is some form of “these are all good ideas, but…”9 Klein, during the book tour, has described himself as seeking “synthesis” with several ideas. Reading this book, it seems like that is less a goal and more a way of being for him and Thompson. They hang around with the tech set so they’ve know what it’s like for something to work—they’ve seen too much and are too smart to not see that their way isn’t working but they’re incapable of fully abandoning their worldview. They can’t stop synthesizing, meaning that they can’t abandon fully any of their allies regardless of performance. Not even their past selves.

When looking at CEOs of the recent past, Steve Ballmer is usually painted in a terrible light. Forbes once called Ballmer the worst CEO in America. Shareholders like David Einhorn called for him to step down. Vanity Fair called his entire term as CEO a “lost decade.” The story goes that he was such a bad CEO that he was paid $1 billion just to quit. His ultra-high energy persona became such a joke that one of the most flattering profiles of his tenure was called “No more Mr. Monkey Boy,” referring to his infamous dance. Microsoft’s stock performance fell during his decade at the helm.

That said, much of the analysis of Ballmer’s time as CEO has been histrionic. We now have the benefit of time, so we can evaluate how he did with more rational eyes, as well as the benefit of hindsight.

So, here is a deep-dive review of Steve Ballmer’s performance—the good, the bad, and the ugly.

Although it would be incorrect to call Ballmer the best CEO of the 21^st century, he may be the most underrated.

Ballmer was good at normal business stuff…

The main Microsoft business did well under Ballmer. He brought revenue from $15 billion to $70 billion during his tenure and earned nearly a quarter trillion dollars of profit while CEO, perhaps unsurprising for a sales-focused leader. He launched several successful and market-leading products, including Xbox, and acquired important strategic ones, like Skype. If the job of a company is to make money and the job of a CEO is to steer the company’s financial performance, then Ballmer succeeded at its most important mission.

Ballmer’s most impressive choice is succession. Most CEOs fail here: according to a Harvard Business Review article, over 40% of public company successors fall behind within 18 months; more still fail to excel at all. Satya Nadella, now the CEO of Microsoft, was merely one of several candidates considered, and he was the dark horse option. Like any blue chip company, they also attracted several outstanding outside candidates like Alan Mulaney, who turned around Ford, and Pat Gelsinger, now CEO of Intel, and among insiders Tony Bates was considered the most likely option. But Ballmer made the smart move: he elevated a technically capable leader who was leading cloud, which was, at the time, Microsoft’s fastest growing sector. Satya will likely go down as one of the best successor CEOs ever. Though Satya’s success is all his own, choosing him as a successor deserves a lot of praise. And getting something big right that most people get wrong makes it especially so.

Arguably, Ballmer did a good job with developing people overall, not just Satya. Cultivating developers was Ballmer’s big thing. It always has been. And it is actually really difficult to do—Apple, Google, Amazon, and pretty much all of Microsoft’s peers have been way less successful, if not outright failures in some cases. Ballmer also created the One Microsoft initiative, creating a more cohesive organization. It was panned at the time by commentators including Ben Thompson. But it worked, and today remains the bedrock of Nadella’s Microsoft.

Notably, this was all done while operating with the difficult aftermath of antitrust. Bill Gates, who remained on the board, famously said that not settling the antitrust case early was his “biggest mistake.” Well, he didn’t have to live with the consequences of that mistake—Ballmer did! And that mistake meant that every new decision was under government scrutiny. Achieving any growth under these circumstances is impressive. In fact, he just generally seems like a reasonably competent manager. Since he bought the Clippers, for example, they have nearly doubled in value and half of their conference finishes ever have occurred under his short tenure. They’re even doing better than the Lakers!

Subscribe now

Despite this, Ballmer was never rewarded in the stock market. Microsoft stock fell approximately 40% during the Ballmer Era. He can’t blame this on the Dotcom Bust either—the NASDAQ only fell 10% during that time. To add insult to injury, IBM even briefly exceeded Microsoft in market cap in 2011.

There’s a reason for that.

…but Ballmer was not a product visionary

Steve Blank once observed that visionary CEOs rarely have visionary successors because they choose operators as successors, and operating and vision are usually mutually exclusive. Ballmer, as mentioned above, was a truly excellent operator—increasing revenue 560% in 12 years is impressive anywhere and is particularly so at Microsoft’s scale. But in several key areas, Microsoft failed to lead because of product failures.

The emblematic failure of the Ballmer era is the iPhone. He missed this, and there’s no way around it. For everything Ballmer did right, this is legitimately one of the biggest modern business failures.

Most damningly, even 15 years later, he still tries to rewrite history. Ballmer has come up with lots of excuses over the years, none of which hold up even a little. It is possible that he still doesn’t understand why Microsoft lost to the iPhone, which would be even worse.

Ballmer originally used to say that his doubts stemmed from the iPhone’s price, and that carrier subsidies were a “business model innovation.” The problem is that this is just incorrect. Phone subsidies were common, including for Blackberry and the Razr. Ballmer has also said that Microsoft was too late to the mobile phone market. This is also wrong—Microsoft had Windows Mobile, and in 2006 it was the most popular OS with 37% of the mobile phone market.

Antitrust has also come up as an excuse, but that whole episode concluded half a decade before the iPhone launch. Years after Ballmer stepped down he was still obsessing over Microsoft’s dearth of hardware products, yet Google’s Android, which is more widespread than iOS, has never had a successful first-party phone.

The reality is that Ballmer was just wrong about what people wanted. The iPhone’s critics saw it as an overpriced, underpowered machine using an overcomplicated OS based on desktop-class software. Ballmer was one of those people. At the time, he thought the iPhone needed a keyboard to succeed and said so contemporaneously during his famous news conference following the unveiling. Buying Nokia was the perfect proof he didn’t get it. A total waste of money in chasing where the ball was, not where it was going.

The other huge product failure for Ballmer was the OS. Ballmer continued to focus on Windows as the lynchpin of Microsoft’s strategy because that was what had always been its forte. While OSes remain strategically important, they are not commercially important. Despite that, while Apple kept decreasing the cost of OS X until it became free in 2013, Ballmer kept trying to get people to pay more for Windows. When Ballmer stepped down, Windows had already begun a rapid decline in global OS share. Android surpassed it in 2017, in large part due to being open source.

This was very much accelerated by the failure of Windows Vista, the biggest desktop release under Ballmer’s tenure. Vista was buggy, it had security flaws, it was slow, and it reduced battery life, which was particularly bad during a time when consumers were buying more laptops and needed more efficient processors. Developers, long Microsoft’s biggest constituency, hated building for it. The criticism was so intense and universal that Windows Vista criticism has its own Wikipedia page. It is one thing to fail so badly that you precipitate a massive loss of market share in a previously-successful product; it is another to do so in a product that you yourself consider important.

As a general matter, this is where Ballmer most critically failed: he was not a product leader at a time when Microsoft needed great change. Ultimately, this is why he couldn’t stay as CEO.

However, Ballmer’s choices set Microsoft up for success, legitimately…

Though Ballmer was a terrible product mind, his business choices set Microsoft up for success. You could even argue that Microsoft still stands on Ballmer’s shoulders.

The most important example of this is Azure. Nadella is, rightfully, praised as the visionary and driving force behind Azure. That said, Azure started under Ballmer. In fact, it began with a Ballmer acquisition, Groove Networks. And as CEO, Ballmer could have killed Azure at any time. In fact, most CEOs would have. The pressure to do so must have been intense because it meant going against every instinct that Microsoft had built up to make it into the most valuable company in the world: selling ultra-high margin software while avoiding cannibalization and hardware. Not only that, but Azure initially stumbled because it was offered under a Platform as a Service model, not Infrastructure as a Service. How many CEOs would have had the vision and courage to stick through with it, especially under criticism from a board that already wanted him gone? Nadella himself praised Ballmer for supporting Azure in its early days. He deserves immense credit for this.

Ballmer also pushed for new products that led to Microsoft having the most diversified revenue of any large tech company. Bing, for example, was a success. Though Microsoft didn’t unseat Google, Bing is now extremely profitable, with revenue exceeding $8 billion, over half of Windows’ $15 billion—in fact, if you include LinkedIn’s $10 billion in revenue, Microsoft now makes more in advertising than it does from Windows! Ballmer also personally kept Microsoft Research funded despite high shareholder pressure to stop. His investments in Office and Azure allowed for the launch of Office 365. Today, Microsoft Research’s innovations have powered many of their most popular and futuristic projects. And the distribution machine Ballmer built is what has allowed Azure to have a real shot at unseating AWS instead of GCP.

Those investments positioned Microsoft to be the only credible company that could back and save OpenAI. It is the apotheosis of investments from the Ballmer era that bore fruit a decade later: it would not have happened without the cloud compute of Azure, without the distribution of enterprise sales, without the data crawled by Bing, without the staff at Microsoft research, without the GPU relationship with Nvidia that started with Xbox, and more.

Putting this all together, we can see how Ballmer set up Nadella for success. It has set Microsoft at the forefront of the AI revolution as the first- or second-most valuable company in the world, depending on the day.

…though Ballmer held on to the past too long

Ballmer has described himself as loyal. It’s therefore fitting that some of his biggest mistakes involved holding on too long. Primarily, everything Ballmer did was about supporting Windows instead of evolving Microsoft. There were times where holding on was a good thing, like with Office, where fighting hard to maintain Microsoft’s leadership gave it a powerful flagship product to power through the cloud era and push its ecosystem with Teams and more. But there were other times where it hurt Microsoft greatly. After all, Ballmer’s obsession with the 90s-era Windows-centric strategy also meant that he did not see the threat of AWS quickly enough, giving them a seven-year head start over Azure which Jeff Bezos called a “miracle.

Microsoft had a defensiveness in its culture stemming from the beginning from its need to convince people that software was even worth paying for, going back to Bill Gates’s famous letter to the Homebrew Club. It is therefore no surprise that Ballmer called open source a “cancer.” In fact, Satya’s embrace of open source is probably his biggest break from Ballmer. Though closed source worked for the 90s, by the 2010s the future was open source. We don’t know how Ballmer would have adapted because this was after his time, and later in 2016 he was singing open source’s praises. But had Microsoft continued down Ballmer’s path it would have been doomed to a slow, irretrievable death given how important open source became to enterprise software. Although it is possible to imagine Ballmer buying Github, it is impossible to imagine him open sourcing VS Code.

Ballmer’s “bull dog” personality also caused him to hold on to outdated management practices for too long. Gates was notoriously difficult to work for and created a competitive culture at Microsoft. Ballmer was no different, and he was most famously a proponent of the “stack ranking” system where the bottom 10% of all engineers were fired every year. This lead to a culture of fear and an average turnover rate of 8%. Unsurprisingly, when Ballmer stepped down, it was the first thing to go. He kept this system despite implementing One Microsoft, meaning that Ballmer’s management style record must be viewed as flawed.

So why the Ballmer hate?

Some of the hate comes from Ballmer’s personality. Ballmer’s flamboyant personality makes him an easy target, but he’s a smart guy who majored in applied mathematics at Harvard and outscored Bill Gates on the Putnam exam. His intense persona is part-real, part-act to motivate employees, and besides, his over-the-top performances were objectively less silly than those of Jack Ma. Ballmer overcame intense childhood shyness; his yelling was probably overcompensation and, seen in that light, actually admirable and endearing.

Yet more hate comes from the fact that Ballmer was a business CEO sandwiched in between two technical CEOs in an era where people are skeptical of nontechnical leaders, at least in the tech community. Ballmer’s predecessor is a genius programmer who was the richest man in the world for over two decades; his successor is an inspiring engineer whose technical vision is somewhat of a rewrite of Microsoft’s modus operandi. It is hard to not look bad.

But more than anything else, it is product. Microsoft was seen as incredibly scary and dominant and it just completely whiffed it in the most visible and embarrassing way possible. Even Paul Graham once said that Microsoft was dead. Microsoft’s mobile position is completely unrecognizable relative to its position in all other computing verticals, it didn’t win the #1 spot in cloud computing that it should have had, and it almost missed open source. Even if Ballmer did everything else right, such big product misses would exclude him from ever being considered a truly great CEO.

But being the best isn’t the same as being underrated. Most people act like Ballmer got a report card of straight F-minuses. But in truth, as a student of business, he received a mix of A+s, Fs, and normal Bs and Cs. A pupil with extreme talents who had clear weaknesses and wipe outs. A 3.3 or 3.4 GPA hiding extreme strengths and weaknesses. Interesting, and not worthy of only being judged by his best, but far better than judging him solely at his worst like most people do now.

So was Ballmer the worst CEO ever? Clearly not. He was the personification of a dribble drive motion: he might have taken some hits, and he might not have been the star, but he set Microsoft up for layup after layup.

Subscribe now

The Vision Pro is really exciting. I can’t wait to get one. Palmer Luckey, founder of Oculus, presciently said that before something like this “can be something everyone can afford it has to be something people want.” Mission accomplished.

But it’s also overengineered. Mark Gurman say that Apple’s launch strategy is taking a “scattershot” approach and letting the market decide what it’s for. App developers aren’t tackling it yet, and the Vision Pro will launch with a smaller App Store than the iPhone or Apple Watch.

You see this problem throughout the products space. Designers are afraid to commit to certain use cases, lest they be wrong. Business planners fear not being able to cover a vertical. So no one has an opinion and every product can do everything, but only somewhat.

The great virtue of the Apple of old was focus and its predicate: choice. To make product choices, you must have a vision of what the product must be. That allows you to take advantage of nascent and amazing yet immature technologies to appear on the cutting edge.

The Vision Pro is an example of abandoning that approach. It wants to be two different products. Let’s explore.

iChoose

Apple went through two distinct periods, technologically, in its history: a Wozniak and a Jobs period. In the beginning, Apple developed myriad technological breakthroughs itself, many from the genius mind of Wozniak. The Apple I and Apple II were products of revolutionary engineering. The Mac took half a decade of hard invention, from the hardware to the GUI. This approach continued into John Scully’s tenure. Products like the Apple Newton and LISA, although commercially failures, involved home-grown technical wizardry on the hardware side. Apple even used to have its own factories in Cupertino.

The Jobs era was extremely different: the modus operandi was using existing, cutting edge technology that was on a steep growth curve and using design and use-case opinions to exploit it while working around the limitations of the technology at the time. Although the post-NeXT Apple had numerous technical innovations, none of the major breakthroughs came from Apple itself. Like a magician, this allowed Apple to redirect the attention of users away from the shortcomings of a technology while appearing cutting-edge by taking new tech that was obscure and bringing it to the mainstream for the first time, often used in a novel way. They took products where free wasn’t cheap enough but with technology that was on the bleeding edge to creatively design products the masses clamored for.

Let’s take two examples.

The original iPod ad, “1000 songs in your pocket,” highlighted its two big selling points: a large music library that was small enough to carry. Before the iPod, you got (at most) one of these. Apple’s major idea was that hard drives were getting much, much smaller, so you could fit 1000 songs on a single device. But Apple didn’t invent a better hard drive; it bought the smallest one it could find from Toshiba. The engineering efforts to make the technology work around the hard drive were immense. But because the use case was incredibly clear, everything else was able to be sacrificed. You couldn’t play videos because the processor was very underpowered and the 160-by-128 pixel screen was bare-bones. After all, the only thing the original iPod did was play music; it was practically an accessory to a Mac. Such simple functions required a new interface, which was found in the iconic scroll wheel (which was made externally, of course, by Synaptics). And because the iPod played music so well, no one even asked whether it could do anything else. Besides, time was on Apple’s side. With Moore’s Law, it wasn’t long until you had an iPod with a vibrant color screen that could even play video games.

The iPhone has a similar story. Apple didn’t even invent the pinch to zoom interface; it came out of Bell Labs. The actual tech was from a company called FingerWorks and the screen was manufactured by Corning. Apple’s main contribution was the software. It was basically a giant screen with big battery and a similarly big price. The battery, which had 1400 mAh, was nearly 25% larger than the 1150 mah that was standard. Despite that, the processor (which was made by Samsung; the P.A. Semiconductor acquisition was in 2010) still had to be underclocked by 33% to support acceptable battery life. The first iPhone didn’t ship with 3G and wouldn’t let you change your wallpaper! With these limitations, Apple couldn’t take the risk of allowing unlimited customizability. (You probably couldn’t do much as a developer, anyways.) Before “there’s an app for that,” there were only a few apps made by Apple. But they had a clear view of what the most compelling use cases would be—maps, texting, the web, video. So they wrote all of the first software. And because it was so good, no one even noticed.

What is key to both of these stories is that Jobs had already decided what the products would be used for in their first iterations and had a design that would make them magical. The limitations allowed him to act like a magician: he could point them in the direction of the amazing new technology while redirecting their attention from the limitations of technology as it stood at the time because it simply couldn’t do, or didn’t need, what it didn’t have. Then, as the technology steadily improved, he added in what was missing to end up with the products we know today.

This also helps visionaries explain their technology. When the iPad first came out, many commentators weren’t sure what it was for. But Jobs had a very clear view: it was the “in between” device. What did that mean? Leisurely reading and streaming. He literally put a living room chair on stage and sat on it, using that setting to demo made-for-iPad apps like iBook and the New York Times. If Jobs had thrown tablets at customers and said “you figure out what this is for!” it would have been difficult to grok. Being opinionated early on made the iPad easier to market, even though its uses have expanded over time. It’s the complete opposite approach to what Apple did with the Apple Watch even though both products were amorphous to consumers at the beginning.

Defining your use cases meticulously also allows you to have favorable price positioning, even when you are selling a premium product, because you get to define your competitors. When Jobs introduced the iPhone, he gave a masterclass in positioning. He immediately defined the phones that were his competition—conspicuously, the most expensive phones made by each of his competitors. Then he defined the iPhone as three products in one—not a product, a bundle of products. So $700 for a phone? Very expensive. But compared to the most expensive Blackberry, and an iPod, and a mobile browser, which would normally require a Mac? That made the iPhone’s premium prices seem like a good value: you were paying a lot, but you were getting even more. If Jobs had just introduced the phone without defining the use case, consumers and journalists would be comparing the iPhone to flip phones without even understanding the difference.

Lacking Vision

That brings us to Apple Vision Pro.

Apple’s new Vision Pro looks amazing. But it clearly came out of a different process than the iPod and iPhone. It didn’t take 1-2 years to create; it took nearly a decade. It’s overengineered. And yet Apple still hasn’t defined what it’s supposed to be for. It’s trying to straddle all levels of the spatial computing experience instead of picking specific use cases and building for them.

You hear a lot of reviewers say that “these aren’t really the devices we want” so all headsets are beta products. But what if it was possible to have those devices today by being smart about the product?

What I propose is that we can have it all, just not at once. Apple, like everyone else, is trying to cram two products into one and having to embrace too many trade-offs as a result. Instead, it should release two products. I call them Air Vision and Mac Vision.

Air Vision: The HUD

Think about pop culture representations of smart glasses. Minority Report, Free Guy, Her. They don’t look like Magic Leap, where you have complex graphics overlayed on the real world for games. They look like an avionics HUD. They provide subtle changes to the real world in a way that looks like a video game. The HUDs of our dreams help us get around the world with labels.

If you don’t need complex graphics, the technology is actually already out there to display simple information in regular-looking glasses, like the Intel Vaunt from 2018 and the XReal Air 2 from 2023; even Google Glass, which came out in 2013 (!) had sufficient graphics for a HUD. What has changed is the availability of on-device machine learning that would allow for dynamic, interactive, contextual graphics. In fact, if it paired with a phone, you could even offload the heavy lifting for harder tasks and create an astounding experience. You’d just need to truly narrow down the use cases. By only doing very simple graphics and a limited number of tasks, you could plausibly get all-day battery life. And now that there are electrochromic glasses like the Ampere Dusk, you can use the glasses indoors and outside with the press of a button, so you can also take it everywhere. Apple has also built awesome directional audio into the Vision Pro, so you could imagine using that instead of requiring AirPods (though of course you could pair it for more privacy; another Apple Silicon home court advantage).

The way I see it, there are four killer features, and amazingly, they are all doable today:

• Live navigation

• Live translation

• Object identification

• Visual aids (like presenter notes)

Not only are they achievable, they’re easier than ever to do. They are also, frankly, applications that work just as well with an accessory to the phone rather than a replacement. The AirPods have proven that there is a $20 billion market for accessories that work well with the iPhone. And since no one wants to take pictures with their glasses, you not only have removed a battery draining feature, you’ve removed the social Glasshole stigma that accompanies products like the Meta Ray Bans. How can you be a jerk for recording people if that isn’t even a feature?

This is also a product Apple is uniquely positioned to make thanks to Apple Silicon: they could create an ultra-efficient chip with a machine learning core, an ultra-low latency communications chip for connecting to the iPhone, and offload the heavy tasks to the iPhone like it does with AirPods. It’s like the iPod: taking something super cutting edge and only using it for a few narrow but awesome applications.

Imagine you’re at a conference, and wearing your Air Vision glasses that paired with the iPhone and AirPods. You could initiate navigation from your phone or by voice with your Air Vision. Instantly, your glasses use an edge machine learning model that detects objects to help you navigate with contextual arrows that point the right way using Apple Maps. For indoors, they use the tech designed by Hyper. If someone talks, you can have the language translated instantly to your language but in their voice with AirPods while signs are translated by the Air Vision. On the way you see an unfamiliar but eye-catching skyscraper; the Air Vision recognizes that you’re staring and pulls up a 2 sentence description, letting you know that it’s a local landmark. Later, you present with Keynote. You don’t need to look down for your presenter’s notes. They’re right there on your face. At the end of the day, you go to a basketball game. Your Air Vision asks you if you want to display live graphics; you say yes, and while you’re in the stadium you get your own personal Jumbotron. When you leave to get snacks and go to the bathroom, it’s smart enough to change from a Jumbotron to a small score tracker so you don’t miss anything.

There is a version with regular glasses (with or without prescription) and a version with sunglasses. They each cost $500. They don’t work without your iPhone, but neither do your AirPods, and you love them. Of course, you buy both.

Mac Vision: The Anywhere Desktop

Let’s contrast this vision with the futuristic representations of immersive spatial computing. Ready Player One. Snowcrash. These are ultra immersive devices that entertain you. They make you perceive depth. Left unspoken is the idea of work. We need more and bigger screens, yet they take up more and more space, and then when we travel we have the same 15” screen we’ve always had. What if you could have infinite screens without needing a giant desk and you could take them anywhere?

Whats the solution? It’s the Mac Vision.

The Vision Pro is trying to cram in a HUD, but you can’t take it anywhere. The battery life isn’t good enough to use for extended use cases and it’s too heavy to wear comfortably for many hours. It’s too delicate for travel and is missing a lot of key features. Somehow, they shipped a work machine with a cruddy keyboard and no hot-swappable battery. The reason is that they didn’t realize the Vision Pro isn’t the successor to the iPhone; it’s the successor to the Mac. What it offers is the unification of the desktop and laptop experience. That’s never before been possible before but it would be amazing. And like the iPhone, you’re getting a bundle of devices: the most capable Mac ever and a breakthrough portable screen that gives you as many as you want.

With that in mind, they can ditch the expensive Eyesight. That will cut weight and cost dramatically while making it more durable (and therefore travel-friendly). Ship it with an M3 processor so it can truly replace a MacBook (it should have never been a generation behind; if anything, it should be the first product to get a new M Series to drive sales). Introduce a special communication protocol between Mac Visions. Produce a special, portable Mac Vision keyboard and make a sleeve on the case that can hold the Vision Keyboard. Don’t worry about recruiting game developers; games take forever to make. Just get enterprise app support. That’s easier anyways and can mostly be handled with Mac app support and a good browser. And for entertainment, a few key sports partnerships is enough. Make the battery bigger so it lasts 6 hours, especially now that you don’t have to power the Eyesight, which probably uses a lot of power. Since people will work all day, make the design ergonomic even when plugged in. Don’t use heavy aluminum; use a thin aluminum coating over plastic to cut the weight (and cost). And push the Vision video chat “persona” available by SDK to everyone from Day One: Zoom already supports it, but everyone else like Microsoft Teams should too to normalize the concept and maximize business user appeal. Build some truly immersive built-in apps—what if you place yourself in Apple Maps street view and travel anywhere in the world for free?

Imagine you’re on the train to the office. You pull out your Mac Vision. You open Slack in one window, Word in another, and Salesforce in a third. You get an iMessage from your daughter, which you answer. It’s amazing; you have almost no space yet you have 3 concurrent monitors open. With pass through, you’re fully aware of what’s happening around you and don’t miss your stop. You put your Mac Vision in its slim hard case and go. It’s smaller and lighter than your MacBook ever was. At work you pick up and resume. You couldn’t get budget for a second screen before. Now you have as many as you want. When it’s time for lunch, you plug in your battery pack and walk off. On the way home, you pull up Apple TV+ and watch an immersive show you love streaming, like Silo, with spatial audio synced to your AirPods Max. The episode finishes just in time. You’re plugged in on the train so you don’t run out of battery. Once you’re home, you’ve gotten 10 hours of use, work and entertainment.

It costs $3000. But your MacBook Pro was $2500 with the same processor and just a 15” screen. The battery life is lower, but you never used your MacBook for more than 5 hours without plugging it in anyways. You happily replace it with a Mac Vision.

And as the technology improves, the problems will start to get better. The Mac Vision will get smaller with better battery life. The Air Vision will be able to display more advanced graphics. And one day, these products will likely converge into one.

Defining the Vision use cases by breaking them up also short-circuits one of the shocks for the Vision Pro: the sticker price. Apple didn’t define its market so the press did it for them, and they chose Oculus devices, which are 5x cheaper. Splitting the Vision Pro into the Air Vision and Mac Vision would eliminate any comparisons to such comparisons because both feed into different markets. The Air Vision would be viewed as a pairing with the AirPods Pro, which cost $250. Sure, $500 Air Vision glasses are more expensive, but not exorbitantly so, and in fact that is the same price as the AirPods Max for a sci-fi product. You’d be delighted to spend $500 for that product, even to spend $1000 for regular glasses and sunglasses. Similarly, the Vision Pro doesn’t seem so expensive when you compare it to the cost of a Mac Studio with two high-end monitors, which could easily cost over $5000. But with Mac Vision, the comparison is explicit. It’s a MacBook replacement. And considering that MacBooks now regularly exceed $2000-3000, you’d feel like you were getting a bargain for a Mac Vision. A computer with infinite screens, and you can take it anywhere, and it’s just the price of a low-end MacBook Pro? Sold.

Despite that, you’d end up with a better price position while actually getting customers to spend more. I think most people would have bought both an Air Vision and Mac Vision. In the end, the customer would have still spent $3500 but spread across 2 devices. Each would purpose built and feel like premium value. Customers wouldn’t even notice.

Choices make products

Making choices was the only requirement for these exercises. These products are both possible to build today. The Mac Vision is just the Vision Pro with unnecessary feature bloat cut away. And the Air Vision is just the Apple version of smart glasses that already exist. Apple is uniquely positioned to offer these products. They make the world’s best chips and can integrate their products to buttress each other.

This future is possible today. I hope Apple gives it to us.

But more importantly, the essence choice is to allow you to build the future by pulling forward only the most crucial features at first. The most cutting edge technology is already out there. It just isn’t fully developed. Thinking about what that technology is good and bad for allows you to map that onto the problem space you are working on. As long as the technology is rapidly improving, you can add the missing use cases later and even anticipate them. You’ll always seem ahead of the curve.

What works for products can also work for venture. In a 2010 talk at Stanford Business School, the founder of Sequoia, Don Valentine, explained the simple reason that Sequoia seemed to have seen the future. He had been the head of sales for National Semiconductor, and he was a believer in Moore’s Law, so he had spent five years mapping out the next several decades of the computer industry. He knew what people would want and what the gaps were. Mapping that out and comparing it to the state of the art then became a marketing exercise.

What you cannot do is give up on thinking. The reason you might want to avoid making a choice is to avoid risk and avoid the difficulty of thinking things through. But as in many things, not doing the hard part and running away from risk is often the riskiest thing of all.

The Silicon Valley Way

There are two main virtues to this approach. The first is to focus all efforts on technology and product. Technologies may be extremely complicated, but in the Silicon Valley Way, that is it. The reason is, true to form, simple to understand. The focus on the total simplicity of business allows for all of the company’s energies and investment to be towards the technology itself—time not spent arguing with the accountant is time spent in the lab. With early companies in Silicon Valley history, like Cisco, there was a maniacal focus on the technical problem.

The second is to force clarity of thinking by avoiding the shroud of complexity. In a recent interview, Doug Leone, the former leader of Sequoia Capital and a veritable dean of Silicon Valley, emphasized the importance of simplicity in the Silicon Valley model. With the best companies, he notes, “it’s never something you scratch your head, you have to go to your [] expert and say, what does this mean? It's never that.” In one crazy example, Paul Graham of Y Combinator invested in Benchling for the dazzlingly simple reason that because biology was important, making their most important software would be valuable.1

This simplicity applies to even the most mundane aspects of the business. Tech companies all use the same lawyers, who use the same form documents from favored institutions like NVCA to create the same legal structures, all of which are incorporated in the same state, Delaware. They use the same rule-of-thumb stock option pool sizes, with standard vesting schedules, and of course all offer equity to their employees (at least, the ones in the corporate offices). They describe their business models with the same acronyms and calculate their financials across the same few common summary statistics. Though this is rarely talked about explicitly, this code is quite rigorously enforced by a lack of funding for startups with the temerity to vary.2 In fact, venture capital as a whole operates basically the same as the days of Arthur Rock.

This is not to say that Silicon Valley avoids business innovation per se, just that those innovations are simple. Free websites supported by ads were novel, Bill Gates transformed software,3 SaaS turned tech on its head, and founder shares turned the Valley “founder-friendly.” In fact, the story of Google is one not only of technical dominance but of brilliant business realizations, like the importance of being the default search engine and auctioning ad space. But these innovations are simple. They involve a straightforward value to the customer and a straightforward exchange of money. Though the act of coming up with these models required creativity, once imagined they are easy to comprehend.

We’re witnessing the suffering caused by adventures in governance now with OpenAI. Like many generational companies, OpenAI’s birth involved changes to its structure born out of concerns that are not crazy. The people who work in AI tend to be true believers and have talked about AI being owned by the world for a long time—even Larry Page once talked about donating true AI to the United Nations, and nonprofits are allowed to own for-profit companies (they’re just usually cafes or bookstores). The idea of creating a corporation that is legally bound to a mission, as nonprofits are, was an attempt to tie the company to the mast of moral obligations when the sirens of profit came calling. This might have worked if not for one problem: nonprofits have no mechanism that forces their boards to behave seriously. There are no shareholder lawsuits, no SEC investigations, not clear metrics for what a “mission” is. And so, OpenAI got what nearly every nonprofit board has, which is a board of misaligned, unserious people.

Though this concept may seem obvious, it is in fact not true for a number of businesses. Finance is a classic counterexample. Most of finance, at its core, comes down to a simple disagreement: two entities look at some type of asset, or collection of assets, and disagree as to its worth, and bet against each other in various ways. The resolution of that disagreement is where the mind-twisters begin, and it shows that finance is just plain complicated. It can require a PhD just to understand some of the valuation methodologies. Something like trading options using Black-Scholes equation is not something the average person can understand and participate in; a Bing ad auction is. Bill Ackman, a billionaire investor who once published a diligence spreadsheet so large it could crash a computer, has called bank balance sheets “inscrutable”—who has ever said that about, say, Cloudflare?

Where a decade of businesses went wrong

In 2011, Marc Andreessen famously said that “software is eating the world.” This mantle, which for a decade was the slogan of his eponymous venture capital firm, was taken up by a wide variety of startups. They took to using software to attempt to create novel, differentiated businesses that rethought existing industries using software. Many of the 2010s-era tech companies were not a total rethink, like Amazon was for bookstores; rather, they had much in common with the industry they sought to replace, like Root Insurance. These businesses were later dubbed “tech-enabled” businesses, 

Andreessen’s essay is mostly remembered for its general idea and catchy title, but the actual text is much more nuanced and circumspect. His excitement was driven by the massively decreasing cost of cloud computing and the potential for the internet to allow for scalable distribution, but he is far more nuanced about the ability of those companies to create differentiating value necessary to maintain high valuations. There is also a strong flavor of the rise of the importance of the digital domain and of software commoditizing its complements. In other words, in Andreessen’s original essay, software eating the world is not a given—it is something that occurs for a specific set of reasons. Unfortunately, many tech-enabled businesses did not meet those conditions. There are two main ways these new tech-enabled businesses flew off the rails.

The first is that tech enabled businesses were all about using new technology in old industries with a bet that they would make traditional businesses look more like tech when the opposite turned out to be true. Technology may change the operations of a business, or provide a temporary edge, but if it is ultimately the same service or product, competitors will become wise to the edge and compete the new advantage away, turning a new technique into table stakes that must either be spent expensively acquiring market share before the market structure stabilizes or absorbing differentiation profits until they are commoditized. There are times when technology is so transformative that it fundamentally changes the economics and dynamics of an industry, like with many marketplaces, and so the bigger the transformation the more deceptive it can be. New companies like Uber and Lyft could only exist due to a confluence of technologies—3G mobile, GPS, and touch screens—and so they believed that they were technology companies happening to transform mobility. In fact, so deep was their belief that they both formed self-driving car units, both now disbanded. Yet, like taxi cabs, car drivers remained their biggest expense, and the capital necessary to scale those businesses is much less accessible in a non-ZIRP world. Technology when applied is ultimately a lever for a factor of a business, but when that factor turns out to not be the most important one, that energy may ultimately be misplaced.

The second is that many tech-enabled businesses, particularly in finance, took on balance sheet risk, and the tech companies either did not realize they were doing so or believed that they could create a better underwriting model but, in fact, couldn’t. Much of today’s technology is implicitly assumed to be information technology, and balance-sheet intensive businesses, many information technology companies leverage their edge in the new field of data science to make better statistical decisions. Companies like OpenDoor and Affirm call themselves tech companies, are founded by technologists, and are backed by technology-focused venture capitalists. But they do not actually sell technology. They use technology to make underwriting decisions, for which they deploy capital. Many new so-called technologies are just businesses that were well-timed to use the Internet as a distribution mechanism without having to also build many of the new technologies. And unfortunately, it turned out that these underwriting decisions were either not better, or were poorly applied. And in fact, for long-time experts, many of these mistakes were predictable—for example, Sam Zell observed that WeWork’s business model was not new and had always failed.

Much of the failure comes from a lack of engagement by tech enabled businesses with the underlying industry.4 Take the example of real estate. In the past ten years, Silicon Valley has gone deeper into the real estate game, funding companies that tried to rethink every part of the way we use our physical space. Some, like Airbnb, turned out to be wildly successful; others, like WeWork, became punchlines. Real estate focused startups like to emphasize that real estate is a multi-trillion dollar industry. All of these companies claimed to be leveraging the internet, using technology to get better operations, and even using data science to pick the best buildings. The problem is that in real estate, none of these are where the real edge comes from. The main determinants of the performance of a property, or even a property portfolio requires hyperlocal knowledge and macroeconomic luck that is out of an investor’s control.5 Any such edge would be small and better off sold to many real estate companies as table-stakes SaaS—Appfolio, not Flow.

The funds that manage billions are good at managing their operations, sure, but they are even better at managing their taxes. As a result, successful real estate investors become prodigious producers of paper. A few years ago, WeWork tried to go public for the first time, when Adam Neumann was still in charge, and they suffered from an interesting criticism: it was too complex, and certainly for the tech industry its structure was byzantine. But industries like real estate, or even pharmaceuticals, where master LLCs use spin-off C-corps for individual pharmaceutical assets, complex structures like Up-Cs are not uncommon. Certainly not in real estate, where such optimizations are essential for maximizing returns. Real estate’s privileged position allows it to defer certain tax gains and eliminate others, but only with great structural care and use of the appropriate financial incantations, all delicately and intricately balanced to also navigate countless local considerations. And with millions of real estate owners, there are thousands of them in every Congressional district, so these are not going anywhere.

To that end, why did Silicon Valley focus so much on tech enabled industries over the past decade? The Thielian interpretation is an act of desperation in a world with no new ideas, no technologies that are actually new and worth deploying, an attempt to extract rents in a “tech”-fueled arbitrage like the specious financial machinations of conglomerates of the 1970s. But this is hard to square with an era that produced the foundational research for today’s artificial intelligence, many novel innovations in areas like distributed computing, and the cloud. Mostly, software entrepreneurs heeding the clarion call for software to eat the world have been experimenting, seeing how far their inventions could go, and embracing the utopian view that their magical code could make the world a better place. Modern software was truly novel for many industries 15 years ago, and in a zero-interest rate world, no one was asking whether the productivity improvement was worth the cost of capital.

A fork in the road

Over the past fifteen years, the shift towards tech-enabled business meant that a significant portion of the technology industry moved away from the Silicon Valley Way. But there are signs of a shift back.

The first shot fired in the march back onto the road of the Silicon Valley Way has been the new rise of large language models, convolutional neural networks, and other related technologies. With the past decade’s advances in machine learning, there is no ambiguity as to its value. Machine learning has allowed for useful advances from better translation to image recognition for the blind. Some have calculated that ChatGPT, even with its weird name, is the estimated to be fastest growing consumer application of all time (though it is too early to calculate retention). As Leone would say, you don’t need to call an expert to understand that. To be clear, there are real debates remaining—whether RLHF is the best feedback, whether mere token prediction is enough for many applications, as well as debates as to where value will accrue in the technology stack. But these are technical debates amongst technologists, which is evidence of the vitality of the approach; there is no debate that artificial intelligence would be a breakthrough. It is instructive to compare AI with crypto, though this comparison is admittedly overdone. No one is telling anyone to “have fun staying poor,” grasping for use cases, or running Ponzi schemes—even children understand what AI does. The business models imagined for new AI applications are simple payment for service. And the approach is familiar. It involved a few hundred engineers raising staged venture capital, incentivized with stock options, and working on a focused problem.

There is a groundswell of progress in other areas, too. Many of the most exciting areas of the technology industry today are in areas involving technical breakthroughs. They include new semiconductor architectures, novel battery formulations like lithium air and solid state batteries, large-scale rockets like Starship, quantum computing, a myriad of biotechnology advances like mRNA therapeutics, gene editing, anti-obesity drugs, and bioforges for new manufacturing, and maybe even nuclear fusion. Even Tyler Cowen, who coined the term “Great Stagnation,” predicts it will eventually end and has had to admit we may be emerging from it!

These industries have much in common with semiconductors, which is the original industry the Silicon Valley Way evolved to facilitate. Like semiconductors, they tend to be capital intensive. OpenAI observed in 2018 that the compute expense of machine learning doubles every 3.5 months on average. Cloud companies each spend tens of billions of dollars a year on deploying new servers. Drug development is so expensive, and increasingly so, that in the industry it is jokingly subject to “Eroom’s Law,” which is “Moore’s Law” backwards, to symbolize how costs keep going up. Yet, these new technologies are similar to the historical Silicon Valley technology companies. These technologies all face technical risk and significant upfront development costs, but the resulting industries should all be high-margin with quickly decreasing marginal costs.

In hindsight, it is surprising that zero interest rates did not result in more investment in these types of companies, which suggests that perhaps the conventional wisdom is mistaken. Typically, cash flows are valued at a discount based on the prevailing interest rates. Many traditional Silicon Valley companies involve high upfront costs with cash flows that are far into the future, but with often low market risk. That makes these the types of businesses that one would have expected to thrive in a low-interest rate environment, where they would have functioned as a capital sink for capital searching for a return where cash flows were in the future, but low risk if they passed the technical bar and materialized. However, that is not what occurred. In reality, the zero interest rate phenomenon did not result in an investment in these technologies. Return and risk are inversely correlated, so for low-risk capital to search for high rates of return is to take on more risk. During the era of easy money, low-risk capital flooded to tech-enabled businesses, where there is market risk but low technical risk, even though the pure financial profile (where margins may be much worse when the cash flow streams materialize) is less obviously appealing in a low interest rate world. This suggests that fears that rising interest rates will result in less investment in advanced, capital intensive technologies may be overblown and that in times of low interest rates investors may be willing to take on not just more risk but prefer different types of risks.

Venture investment overall may be down, but companies in AI are raising more than they ever have while industries like biotech have fallen, but not off a cliff. The impact of rising interest rates on true technology may be less than is currently feared. After all, technology is not the same as tech stocks. Venture capital funds are sitting on larger amounts of dry powder than ever, and it is continuing to grow. Though some of this may be illusory, there is still much capital to invest. Before these funds run out, there remains ample opportunity to deploy it in more technologies that comply with the Silicon Valley Way and show returns to continue to raise large funds to deploy in these emerging areas.

Perhaps the greatest risk to Silicon Valley in particular is the separation of design from manufacturing. Apple is famously Designed in California yet manufactured wherever it is cheapest. But moving its manufacturing is no simple task. When it tried to build a Mac Pro, it got “screwed” by a lack of screws—reportedly, the shortage was so bad they had to be brought over in a pick-up truck. It is no surprise that the most advanced electric vehicle batteries that are not science experiments have been designed by Panasonic, Tesla, and CATL—they actually manufacture them. The rapidly decreasing cost of technologies like solar is more generally referred to as Wright’s Law, which is typically explained as “learning by doing.” It is no surprise that many of the best designed drones are made by DJI, which has long been the world’s largest manufacturer of them, nor that many of the best factory robotics systems in the world have been developed by the very factories that use them every day. There is significant evidence that industrial agglomeration effects lead to spillover effects in firm innovation generally. Ensuring the existence of domestic manufacturing, at least for strategic industries, is not just a matter of ensuring supply chains during times of crisis. It is also about maintaining an innovative edge.

As investment decisions are made by the large public institutions, like universities and pension funds, as to continue investing in venture capital, the answer seems to be that greater returns lie ahead by following the Silicon Valley Way and investing in these new, true technology companies. This does not mean that venture capital should not exist for tech-enabled services and SaaS, which have improved people’s lives. But it does mean that those capital allocators should be asking some hard questions. Learning from the past decade, they should be determining whether venture capital partners are truly equipped to evaluate the types of businesses they aim to invest in. They should be asking how they will understand these technologies and better distinguish between funds that invest in true Silicon Valley technologies and those that don’t.

One thing that zero interest rates did do is demolish corporate governance. The revival of this practice will be important for bringing back the Silicon Valley Way. Starting in the mid-2000s, the technology industry became more “founder friendly,” which meant less dilution and keeping founders in positions of power longer. Though this has been good for innovation, it also has a darker side: weakened boards, less accountability for management, more shallow diligence. The main impact is not the rise of fraud but an inability to force startup leadership to focus on achievable goals and true commercial viability. The Silicon Valley Way has always emphasized putting real products in the hands of users and not proliferating useless features or product lines. For the Silicon Valley Way to function, there must be a way to limit these excesses, which will require a revival of better corporate governance. The return of governance, however, must not be a return to the ways of old, which stifled innovators for the benefit of financial investors. Instead, it must be a Governance 2.0, where boards provide true accountability, companies face enough scrutiny to discourage fraud, and founders have the power to see through their instincts and visions.

The Silicon Valley Way is coming back. That said, it is not a given that this future will materialize. The future of technology requires us to make choices and set good incentives. The Silicon Valley Way has been able to keep things simple due to an implicit focus on the end customers and a sense of urgency to make the future happen as soon as possible. The Silicon Valley Way is not a mindset, but it emerges from one. We must emerge from the fog in more ways than one.

1. Design

2. Build

3. Evaluate

The design stage is where you decide what you want to do. The build stage is where you actually create the product. The evaluation stage is more than just testing — it’s deciding, as a company, whether you like the results and whether you go back to the drawing board. If you’re doing a startup the right way, you’re also shipping early, so it will feed into your KPIs too.

In the earliest stages of a company, there is a common division of labor between technical and nontechnical cofounders: these stages end up looking like a sandwich. The nontechnical cofounder only really gets involved in the first and last stage, and the technical cofounder often handles the middle by themselves. Think Jobs and Wozniak: pretty separate roles with different strengths and parts of the process.

Both founders are involved in the design stage because the founders are both involved in the idea. Sometimes, you have a nontechnical cofounder who has some type of subject matter expertise, say a designer or lawyer, in which case their involvement in the design phase is even more critical because they have some type of special insight that goes into the spec.

The nontechnical cofounder is especially critical in the last phase too. While both founders need to be happy with the end product, often the nontechnical cofounder is driving sales, marketing, and other external functions that drive the objective inputs to the evaluation stage. And like before, if the nontechnical cofounder has some type of expertise, their input is particularly valuable in understanding the output and evaluating it.

But the the middle, building, is usually a solo task for the technical cofounder. The lack of involvement of the nontechnical cofounder in the building phase creates issues. The most obvious one is that, especially in the early days, the technical cofounder is even more understaffed and might have to do everything themselves. But most critically it’s a source of tension. The nontechnical cofounder isn’t involved in building, so they fail to develop empathy for what it means to build, making unrealistic asks and getting frustrated when things are harder than expected (especially with unexpected delays; it is difficult to get an intuition for what is “hard”). The technical cofounder sometimes feels like they are working on the only thing that actually matters and develops a sense of superiority, leading to myopia in building and resentment at being told what to do.2 This can boil over, and indeed, cofounder conflict is one of the top reasons early stage startups die.

LLMs change this in an important way: nontechnical founders, for the first time, are involved in the build phase.

With LLMs, the technical founder has lots of new things to do. There are vector databases to maintain, representations to manage, tuning to orchestrate, and more. But there are also prompts to engineer. Unlike everything else on this list, prompt engineering is done in pure English, so the nontechnical cofounder can actually get involved. From a pure division of labor perspective, divide-and-conquer logic alone can get the nontechnical cofounder involved in building.

In cases where the nontechnical cofounder has expertise, it is actually an advantage to assign the prompt engineering to them. They are the ones who are best equipped to translate their knowledge directly into prompts and quickly iterate based on the output. Plus, even the most nontechnical founders can learn at least a little Javascript or Python to create dynamic prompts. Suddenly, you will have nontechnical cofounders coding up full-fledged features alongside their technical partners.

The sandwich is thus transformed into a meatball where everyone is involved in the entire product process. This is better overall. The best founding teams are ones where everyone has at least some understanding and empathy for the entire process. This new paradigm will contribute to a whole new set of nontechnical founders who are more involved, more useful, higher agency, and more informed throughout the product development cycle. That’s better for the whole startup and the cofounder relationship as well.

Paintings, jokes, and other creative acts are protected by an intellectual property regime called copyright. Copyright is the easiest type of intellectual property to get, and in exchange it offers the weakest protection. The basic outline of copyright is this: if you have a creative work, people cannot copy it, distribute it, and so on. You get the right to make the sequels, translations, and so on, which are called derivative works. You can even stop people from adapting it from one format to another or using your work in certain noncommercial contexts like political rallies.

The thing about copyright is that, as the name suggests, it is all about stopping people from making copies of the work; importantly, you cannot copyright an idea. Therefore, you can’t stop people from creating their own creative works, like lists that mention your works or analyses of your creations, which is called transformative work.1 Parodies, for example, are transformative because even though they often involve the use of copyrighted material, they transform the material in a permitted way.

The AI companies are going to wipe the floor with these litigants using copyright law as their towel because it’s basically impossible to argue that machine learning isn’t transformative use. Somehow, you can mix Bo Burnham jokes with Picasso paintings and out can come ideas to help you with your new microfluidics project. If that isn’t transformative, I don’t know what is, and it’s the reason neither complaint even mentions the issue. And as for making copies of the works for training sets, copying alone isn’t always illegal, and any sane court is going to grant that this is fair use. Authors Guild v. Google, the Google Books case, said that the book scanning was fair use.2 Google Books to this day displays literal snippet views of books, which was held as transformative; LLMs produce original nondeterminative text, which is clearly even more so. This is, of course, a very simplified analysis (that is not legal advice!) that is not intended to be at the level of a legal brief, so take it with a grain of salt.

It is deeply ironic that this lawsuit was filed by Sarah Silverman because no category of creator is more dependent on fair use law than comedians, who depend on parody law, a type of transformative use, to have a job. Make no mistake, this class action suite is an attempt to shoehorn having a copy of her book, which they are allowed to do, into a legal argument that LLMs are derivative, rather than transformative, works. It looks both weak and a little insane.

What all of these people really want is a patent. Patents have a totally different set of rights intended to incentivize different behaviors. They are designed for the field of engineering, where people discover and invent technologies, sometimes at great expense, and would otherwise keep their inventions secret. The societal trade with patents is the absolute right to stop anyone from commercially using your idea in exchange for teaching the public how it works in excruciating detail.3 Copyrights are porous because the exceptions are legion, but if you are within the four corners of a patent the only question is how much you’ll be paying.

Most pertinently, patents have no concept of transformative use. If you invent something and patent it, you can “block” people from using your idea even if it is a part of their own completely new inventions. We allow this with patents because lots of important inventions that deserve compensation aren’t products per se. But copyright is brittle because creativity is part of the human condition. We want people to be able to express themselves, and references are an important part of that, which means that that the bounds are highly limited on purpose. The whole point of the OpenAI lawsuits are that the litigants don’t believe that spirit should apply to machine learning.

Is there a way out of this for Sarah Silverman? Sure. Intellectual property is entirely created by legislatures. There is no common law concept of IP, at all, in any country, and no Constitution anywhere imposes restrictions on what those rules must look like. The divisions we have between copyright and patent are conventions we have developed because it is commonly accepted that the creative arts and engineering both deserve IP protection, but have different contexts and incentives that merit different regimes.³ This has been the balance for centuries, but what Congress has created, Congress can change.

Copyright could have yet another exception for machine learning use of copyrighted material that is not fair use. It would just require another law. Congress can govern that exception any way they like. They could make it look more patent-like if they wanted. The basic premise would be that if you are using the copyrighted work to create a software product, or a machine learning product, or something like that, the work would not have the transformative use or fair use get-out-of-jail-free card it would normally enjoy under copyright.

I don’t see this happening, frankly. There is likely not much appetite for changes to copyright in the United States; the last change to the Copyright Act was nicknamed the Mickey Mouse Protection Act and Congress hates Hollywood right now, so you do the math. I don’t think anywhere else would do it, either. Most Asian countries are pro-AI, so they would likely not implement changes to copyright that would discourage machine learning. The EU is very willing to go its own way with regulating AI, and it does have an expanded version of copyright called moral rights, so it is clearly not averse to having a stricter notion of copyright. But modifying copyright in that way would undermine over a century of settled business models and hurt homegrown EU data companies, which they would be loth to do.

The alternative is just regulating machine learning, which might be what Eliezer Yudkowsky wants for other reasons, but that isn’t happening, either. How do you determine what the licensing value of a single work is? There are also so many features now that use machine learning. Who would be subject to the regulation? What if it just makes it impossible to offer those features in certain places? That’s what happened with link taxes and almost what happened with GDPR. Can you get around those rules by just doing the training elsewhere? It just becomes impossible.

And frankly, it would be be undesirable. We want machines to be able to learn from the world, and creative works are part of it. The point of all IP is to allow you to protect the products that come out of your mind, not to stake a claim on the inventiveness of society as a whole. Creating a machine learning creative rights patents would be anathema to the idea of creating new things and learning itself. It would be madness.

There will ultimately be some commercial arrangement that enables creators to contribute to the AI revolution and get paid, but being part of a large corpus that enables LLMs to work well isn’t it. Most likely, as in everything else relating to Hollywood, it will have to do with official branding and merch.

Gell-Mann amnesia is the idea that we often fail to update our credibility scores in response to new information. The effect is named for Murray Gell-Mann, a Nobel Prize-winning physicist. His friend, novelist Michael Crichton, described the effect memorably:

Briefly stated, the Gell-Mann Amnesia effect is as follows. You open the newspaper to an article on some subject you know well. In Murray’s case, physics. In mine, show business. You read the article and see the journalist has absolutely no understanding of either the facts or the issues. Often, the article is so wrong it actually presents the story backward — reversing cause and effect. I call these the “wet streets cause rain” stories. Paper’s full of them.

In any case, you read with exasperation or amusement the multiple errors in a story, and then turn the page to national or international affairs, and read as if the rest of the newspaper was somehow more accurate about Palestine than the baloney you just read. You turn the page, and forget what you know.”

The key concept is that we often receive information that should change our world view, especially about the credibility of a source, yet fail to update. This happens all the time.

The opposite is also possible, though, which I call Gell-Mann Paranoia. The reverse situation is that you receive new information and instead begin to doubt everything else in your mental model, or disbelieve everything from the source. Byrne Hobart points out that Gell-Mann Paranoia, for example, is the reason that traders often cut losses — though the loss itself may not be so bad, it often disproves something deep about the mental model, making the trader fear more future losses with a confidence interval too wide to hedge against, and even worse, that the prior gains might have been pure luck.

Taking the example of the FTX blowup, I don’t believe that LPs are upset at VCs primarily because the losses were so spectacular or because the reputational risk is so great but rather because of Gell-Mann Paranoia: they are now wondering which other marks in the portfolio are wrong. Sequoia put out a memo trying to protect itself, saying that the losses were less than 3% of committed capital and that they had already returned the fund in unrealized gains alone. But with Gell-Mann Paranoia, the problem isn’t the mere $150 million loss, it’s that their investors may be doubting the accuracy of the reported $5.8 billion in unrealized gains; if Sequoia got this one wrong, their estimated other gains may also be wrong, so the fund may have less gains than reported (or even be at a loss) for fundamental reasons that no mere audit or GAAP compliance can resolve.

The scary thing about paranoia is that it is irrational by nature, a form of fearful expectation, and the dangerous thing about expectations is that they can get anchored and then create their own reality. If paranoia gets anchored, it could disrupt trust in tech, which would greatly diminish the entire industry. The high trust in tech is what makes it strong because it requires highly skilled people with lots of opportunity cost to collaborate on ideas that initially sound weird and may have no proof yet. It is what convinces people to take chances and try new things, rejuvenating the industry at every turn; arguably, without tech’s abnormally high level of trust, it would not function.

The history of tech is full of shocking acts of trust. While under the spell of Gell-Mann Paranoia, people will spend more time wondering whether their counterparty is lying to them, and perhaps say no to good-but-weird opportunities. That will mean people not working together, new companies not attracting their first customers, and startups never getting started. Get rid of the safety net of trust and participants will take fewer chances, sapping tech’s vitality. Eric Schmidt only left his public-company CEO job to lead Google because he trusted John Doerr when he promised to find him another job if he failed; what happens if the next Schmidt doesn’t believe it?

There is much to take stock of, and it is clear looking back that the tech industry has made many mistakes over the past decade, but we must beat back Gell-Mann Paranoia to prevent collateral damage. Defeating fear is an emotional journey, not a rational one, that is based in reality and not in reasoning; it requires acknowledging the fear, then accepting and facing it, and finally overcoming it. Flushing out the system will thus require pain — layoffs, down rounds, and even bankruptcies — and reflection as many fields that have mostly avoided skepticism, like web3 or scooter sharing, undergo long-overdue reckonings. But the key is to not give up. The fear will go away as those creating true value reveal themselves, thus validating the evergreen truth that there is still much value being created by technology.

What are epicycles? If you imagine an object orbiting another, you will of course model that as an ellipse. The problem is that if you think the planets are orbiting the Earth and they are actually orbiting another object, say the Sun, over time you will have observations that fail to fit that model. Rather than treat these deviations as disconfirming evidence of geocentrism, blinded by religious dogma, astronomers added mathematical complications to their models of planetary orbits starting with Apollonius in the 3rd Century BCE. These modifications were called epicycles, and they were very convenient for the committed geocentrist — when the model broke down, an astronomer would just add a new epicycle and create an even more convoluted view of the heavens that could last for a little longer.

The importance of epicycles is not the (interesting) history of science but the fact that they describe a common intellectual pattern. We engage in epicycles all the time to preserve something that we want to believe is true, regardless of whether it actually is. Armed with the fig leaf of trying to figure out a complicated reality, we hold on to theories of the world like ideological investments, afraid that by moving on we will devalue the intellectual capital we have expended. So we just add epicycles.

You should not avoid complexity, just unnecessary complexity. Albert Einstein said it best: your theory should be as simple as possible, but no simpler. In other words, your theory must incorporate everything that is true and that has explanatory power; if it reflects a reality that is in fact complex, then so be it. If you discover a new wrinkle, incorporate it. But if your theory keeps failing and the only way to make it fit is to add complexity, eventually that is a sign the core underpinnings are wrong.

The key sign that you’re engaging in epicycles if your theories never get simpler at all, and if threats to your core beliefs are always met not with skepticism but with hostility. The complexity of reality means that your mental model will get naturally more complex as you go out into the field. But if you are learning you will also be falsifying. Some beliefs will get thrown away. Some mysteries will remain open. If that never happens and you just keep adding more nuances just to keep up with preserving your core theory, to feel like everything has a tractable solution, those aren’t details — they’re epicycles. Epicycles turn falsifying evidence into confirmatory evidence by contorting the model to fit the data, creating a truth available only to the priesthood that understands an ever-more complicated view of the world. The hard part of rejecting all of this “evidence” is that you need to be able to see the theory fail to make sense and constantly reject new evidence without the theory itself being updated. To transcend epicycles requires you to actually think.

The greatest ally of epicycles is dogma. Geocentrism lasted so long because priests and philosophers decided that the Earth’s placement in time and space had implications for its spiritual importance. That made accepting heliocentrism anathema on a deeper level than science, which is the reason so many smart people toiled to preserve something whose falsity was increasingly obvious for centuries. The dogma was so strong that literally generations of technological improvement, which consistently produced falsifying data, was met with rationalization instead of rationality. That dogma is enhanced by authority figures, giants who made real progress but with a fatal flaw whose work we rightfully revere frequently at the cost of proper, fair critique. Our heroes only become dangerous when they become shields against thinking because, we assume, they have all the answers; the line between reverence and subservience is often exegesis. We can ignore the truth indefinitely if we want to, and our tool of choice is epicycles.

As proof, epicycles lasted, extraordinarily, for over a millennium. Accepting them often is a means to avoid an uncomfortable truth. Rejecting heliocentrism didn’t just mean rejecting a scientific model. Rather, it meant rejecting a philosophical view of our place in the world held by Aristotle and the Catholic Church alike; it meant disproving epochal knowledge passed down from ancient foundational figures like Ptolemy. Escaping epicycles requires not just openness and honesty but oftentimes bravery and fortitude to stand against conventional wisdom. Epicyclic models often get adopted by whole communities, so rejecting them often means standing alone and accepting isolation.

It is lonely to reject epicycles. The more there are, the more entrenched. But the reward is operating under the correct model and being the only one to go in the right direction.

Crypto economics are complicated (so most web3 profits are probably fake)

I’ve often observed that Silicon Valley keeps the business side simple to free up energy to focus on product and technology; web3 is proving that wisdom. Here are two simple examples. When you have a large, international company that does business in other countries it will inevitably conduct business in multiple currencies; those currencies have fluctuating exchange rates (usually modestly) concerning that company’s reporting currencies. Companies adjust and disclose this, but not in crypto even though most transactions are conducted in cryptocurrency but reported in dollars. According to Dune Analytics, OpenSea, for example, did $5 billion in January 2022, which is 625x its volume in January 2021. That’s great, but during that time the price of ETH increased 4x (6x by November!), and Solana is even more extreme, increasing over 100x in 2021. One wonders, how many web3 companies reported financial results in dollars that grew less than the price of their native cryptocurrency? Similarly, accountants argue a lot over whether a particular money flow is a revenue or a cost (like with payment processors). Web3 has not grappled with this at all because tokens are built directly into the operating model. The Decentralized Wireless Alliance, for example, found that Helium, which raised over $300 million from a16z and claims to have over 10,000 base stations, has “real” revenue of just over $6000. One wonders, how many companies report usage fees needed to operate a network as revenue when it is really a cost? Ethereum’s gas is often described as revenue, but is that right?

So far, blockchain’s brightest future looks like a developer technology, not a consumer product

The strongest of evidence that web3 is here to say is, as Marc Andreessen mentioned, talent flows. A ton of super smart people are throwing themselves into web3. Some of this is the intense flow of money, and some of it is just that the blockchain is an interesting technology. But it seems to me that the “deep” reason is that every developer has had to deal with the pain of closed data. They’ve had to deal with legacy APIs and a datasets that a gatekeeper won’t expose. The idea of an ability to have software and data live on the Internet with no gatekeepers or centralized authority is very appealing to developers. It solves a real pain point for them. Combine that with technically interesting work and you have something worth being excited about.

A lot of the coolest things about web3 have absolutely nothing to do with web3

One of the most exciting things about web3 is the number of companies that are rethinking ownership, incentives, and things like profit-sharing. But you don’t need web3 for that at all. You don’t need tokens or decentralization to accomplish ownership. It’s awesome that web3 has caused people to think differently, but it is frankly unconvincing as a selling point for blockchain. Companies have already accomplished incentivization in a centralized fashion with things like stock options and creator funds. In fact, it’s instructive that the CeFi exchanges have been way more successful than the DeFi ones (and Moxxie’s awesome art project was banned by centralized services!). And the idea that you need blockchain for community, which is older in human history than civilization itself, is a joke not worth further discussion. Maybe web3 can change what community means, which is a real and tantalizing possibility; however, there needs to be more innovation here to accept that.

Tokenomics are completely backwards, signal-wiping, and distracting

Tokenomics are an essential part of blockchains not because they do something for users but because they compensate the parties that secure the blockchain. It exists because of the computing cost and the need to incentivize the expense of that compute (that’s not to say tokenization hasn’t been used innovatively when used thoughtfully). Building a startup is hard enough. So now you have to build a product, solve technical problems, and attract users, and design an economy just to get started? Yeah, sure. Web3 needs to get back to its roots with tokens: use them to pay miners/validators/whatever to keep the service free and then just build a compelling service that only a decentralized service can provide. Because tokens are a way to get rich quick, companies now make tokens a big part of the user experience, like play to earn. But as Aaron Levie eloquently put it, all that’s done is obfuscate whether you have product-market fit. Joe Weisenthal pointed out that during the Dotcom Crash, eBay and eTrade both saw growth in their usage even as their stock crashed because they provided real usage. In web3, so far this has not been the case — OpenSea and Axie Infinity have lost more than half their users in the past few months.

Bitcoin’s failure as a currency is underdiscussed and actually important

Bitcoin has failed as a currency, as has its progeny like Ripple. There are attempts to fix it with things like the Lightning Network (which could work — Lightning is brilliant!) and Web 5 (which is a serious effort), but so far Bitcoin is still traded more than it is used. In the meantime, people have pivoted to describing Bitcoin as both a store of value and a volatile way to make money, somehow. While the hopium may one day turn into something tangible, the fact that Bitcoin has failed at its originally intended purpose is actually really important and underdiscussed. Usually, a pivot is unimportant. Companies pivot all the time. But Bitcoin has birthed a trillion-dollar asset class without finding its use case yet. That is unusual, and Bitcoin’s failure is underdiscussed because the notional value has gone up anyways. The normal explanations of being slow and volatile are both proximate causes of failure, not deeper ones. Criticism in web3 is often met with derision and outright hostility —even Kelsey Hightower, a Kubernetes creator (as good a distributed systems expert as it gets) was pilloried for his analysis. This is particularly important because Bitcoin has sometimes been used as a currency for large or critical purchases in failing government environments, like buying medicine in Venezuela. In fact, currency remains the most compelling use case for blockchain (even a decade and a half later!). A deeper analysis is needed to find the right pivot or maybe realize the initial vision.

On-chain data is necessary for web3’s long term success

Blockchains are not yet able to meaningfully store data on-chain. That means that they are fancy certificates, essentially, pointing to an AWS instance somewhere. That’s just as centralized as before and will eventually be replaced by an on-cloud solution. That’s not high-value enough to abandon things like government custody or trusted auction houses. Similarly, for things like digital art (where NFTs make complete sense and might be revolutionary for digital artists like Refik Anadol), the important qualities (scarcity and object permanence) are only meaningful if they apply to the art itself, not the certificate, which requires on-chain data. Otherwise, the art is just centralized in the cloud and they will custody the certificates too, obviating the blockchain.

So far only the science DAOs are interesting

Decentralized Autonomous Organizations, or DAOs, are rather interesting, and the most interesting use case so far is science because it is the only example that needs both decentralization and incentivization. The centralization of science, from government funding agencies to large universities to large journal conglomerates, has clearly corrupted science, prompting the creation of decentralized solutions like PLoS. But science requires coordination and incentivizing because of the need for many parties to effect certain behavior, like journal refereeing or grant writing, so decentralization efforts in science have been successful but not lived up to their full potential. Therefore, having money and incentives baked in for Science DAOs may be a positive to achieve much-needed decentralization, as opposed to many of the above examples where those qualities were distractions or even harmful. In contrast, ConstitutionDAO was funny and raised a surprising amount of money, but it was always doomed and the aftermath was really messy because no one involved had ever really thought about it.

Sorry, it’s not still early

You can only say “it’s still early” for so long. If you think blockchain is a generational technology and compare its uptake to the historical progression of other huge technologies, then yes, it is clearly early over a 50 year span. But uptake is usually measured by wallets and not use. The Bitcoin white paper is 14 years old now. When ARPANET was 14 years old, it had proven its intended use case — fostering research collaboration — and the NSF funded CSNET to expand the early Internet to all universities. When the transistor was 14 years old, which was already proven as the best technology for computing and radios, it had already overcome its biggest technical problem (by switching from Germanium to Silicon), and had proven essential for a number of military and industrial applications. It’s easy to find people who have traded crypto or tried to make money off crypto through things like play-to-earn. It is hard to find people who have actually used web3 for a sustained period of time. Solana has more transactions than Ethereum, but still, where is the great, exciting application?

The market for blockchain might be really small and no one’s talking about

I saved the spiciest take for last. It is not, at all, obvious that the market for blockchain is large. At its heart, blockchain is just another type of database. When NoSQL came out it was revolutionary. It was a database that was truly optimized for the mobile era, so even though no user ever demanded MongoDB, the applications that they wanted required it, which drove its uptake. There turned out to be a huge market for NoSQL databases, but they didn’t kill SQL (turns out relational databases aren’t so bad) and the biggest money went to the cloud providers. Perhaps blockchain is actually a substitute for incentive systems, in which case it could be a subset of something bigger than any database system ever was. But it is also possible that the use cases are few, or small, in which case the juice might not be worth the squeeze. No one has done more to think about potential use cases than Vitalik, but many of them are still unproven and seem small.

Startups and investors look down on secondary transactions. Secondaries are viewed as lower status because they don’t provide capital to the company, so they are a purely financial transaction — they don’t give fuel for the company to grow.1 Companies have other concerns as well. They worry about losing control of who is on their cap table, they worry about disincentivizing employees from working too hard by giving too much liquidity or making the stock price so dynamic that it becomes a distraction, and they worry about unnecessarily setting a price in anticipation of the next round. There is also always the looming threat of regulation.2 These are all valid concerns, and for that reason all startup shares contain a right of first refusal, or ROFR, to control who gets the stock.3 Top firms like Andreessen Horowitz and Sequoia are doing more secondaries, but it is just to get even more exposure to their favorite companies as they desperately search for more ways to deploy their dry powder.

Despite the validity of these concerns, private companies and their boards should look at secondary sales differently. By tackling these concerns head-on and being thoughtful about them, they can actually use secondary sales strategically to better manage their investors and manage their employees.

First, let’s tackle investors. The most common reason for an investor to be turned away is not that they won’t be helpful but because of fears of concerns surrounding dilution or investor size. Secondary shares solve that problem instantly. Instead of diluting by expanding the issuance, you can just sell existing shares and get this person on the cap table. Because secondary transactions don’t require amending the Investor Rights Agreement, you can also do these transactions off-cycle when someone great comes along. From that perspective, secondary investors are not interlopers but partners. Today it would be unheard of to give an observer or board seat to someone who buys common stock, but if the buyer is strategic, it should be treated like any other investing decision.

Secondary transactions can also be used to set favorable terms. Secondary transactions are usually common stock or early investor stock, which sits at the bottom of the preference stack; the investors who want those stock (1) will be your most enthusiastic investors willing to take on the most downside risk, so they will usually pay a premium; and (2) are lower in the preference stack, so they will set a floor, not a ceiling. In formal financing rounds, secondary and primary can be combined to allow a great investor to come in at buzzy headline number while still having a good blended valuation.4 This can lubricate the deal to make it go through.5 Plus not all secondary sales need be common stock from employees — sometimes secondary sales are preferred stock from past investors, which still achieves those goals.

The other issue is employees. Companies need to tightly control this because premature secondary sales have become a big problem. Especially with founders, some have recently been allowed to sell either too early or too much stock before an exit.6 It is all too common today to see a secondary sale for a company that is still only has the mere promise of scale. The founders of startups like Clubhouse and Hopin sold lots of secondary stock before they’d proven that their business was mature and stable, and not surprisingly the companies have since fallen way behind. At some companies, like Bolt, secondary sales have even triggered suspicions of bag passing because they were funded in part by loans. The solution is not to prevent secondary sales but to limit them. The best way to handle this is to take control of the situation and broker limited secondary transactions. This can be highly motivating to employees who deserve liquidity, and is done by companies like SpaceX7 and, recently, Brex.

Rather than demotivating employees, as long as it is a small amount, the company is sufficiently late-stage, and the employee has sufficient tenure, secondary sales can be motivating. In today’s market, companies remain private for long periods of time and employees do deserve to be able to, say, buy a home.8 Plus, just like with investors, secondary sales can be used to manage the cap table on the employee side by cleaning out old employees. Employees who are underperforming or who have departed have an outsize presence on the cap table; reducing it is a good move. Allowing those employees to right-size their equity packages, or even sell once they leave, is a great way to remove “dead” equity that is not actually incentivizing anybody, particularly if it can be sold to someone strategic.

Realizing that secondary transactions can be used strategically should cause the startup world to view secondary transactions as high-status, even though it will never be the same as a pure primary venture investment. But everything can be used as a tool, and secondary transactions are no different. It’s time to start using them that way.

The best engineers I have met aren’t necessarily the most technically minded. They are the ones who look up from their desks.

Looking up means understanding the meta-problem and considering the nontechnical factors behind technical decisions. Engineers are system thinkers by practice, but the success of a project as a whole often involves a wide variety of factors other than technical choices, which means that engineers are often not looking at the whole system when making decisions.

I’ve noticed three main ways engineers look up.

1. Looking up when picking. For many engineers, choosing the company you work at is often more important for your career than the quality of your work. Similarly, understanding the financial implications of your stock package is often more important than the amount you negotiate for in your financial future. Most engineers are largely ignorant of these types of dynamics and have, at best, a surface-level understanding of the situations they get themselves into. Engineers who look up understand the importance of picking and think hard about these factors when making choices.

2. Looking up when operating. Many engineers are considered good when they can think out the technical implications of their work, and many excellent engineers can think of interesting technical expansions that would improve their product. However, very few engineers actually think about why they are building what they are building. They don’t consider which variables are important and often ignore non-engineering considerations. Their energy is infectious and drives other people to drive forward projects. Engineers who look up seek out non-engineering factors and lead even when it is not in their job description.

3. Looking up at the business. Products solve problems or bring joy. Many engineers are excellent at solving the technical barriers to their problems. But very few take the time to understand the business reasons that led to the problems they are solving in the first place. That means that incentives of the actors at play, what people are actually paying for, and deciding which businesses to even go through are almost always considered as outside the purview of engineers. Engineers who look up ask questions about the business and then build with those questions, and not just technical ones, in mind.

Someone I know got advice from a mentor of his on the difference between staff engineers and principal engineers: principal engineers can be given a task and then scope and execute entirely on their own. You can just press play. While I don’t want to go against this particular individual’s view (who will remain nameless but is highly distinguished), this is not the same as looking up. Self-scoping technical tasks is not the same as understanding that task’s “why”. The ultra-skilled domain experts are still necessary (too many opinions will create a black hole) and worthy of respect; looking up, however, seems rarer. You must be technically strong to look up, but looking up is less about skill than choosing to leave the matrix to ask your own questions.

Every time I have seen a skilled engineer who looks up like this they are successful. Here are three real-life examples from friends of mine (with key identifying details ommitted).

1. Picking. One engineer thought long and hard about their first job out of college from a prestigious engineering university when the cloud was still in the early days. They realized that the introduction of the cloud introduced certain security challenges regarding access and became one of the first employees at a big, now-public company. When they decided to move on to their next adventure, they thought about the new security challenges from the expansion of the cloud and the cybersecurity challenges those introduced. They then became one of the first employees at another cybersecurity startup. At the first company, they learned what CIOs and CSOs care about, became better at building products in a secure way, and learned how to navigate the engineering challenges of scaling technical teams and brought that knowledge to their new company. The second company was acquired at unicorn status.

2. Operating. The next engineer was the first outside engineering hire at a privatized open source SaaS startup that is now a unicorn. The space was not yet settled, with other startups pushing their product based on competing open cores. The engineer asked a lot of questions about when their company’s open core would come out with its major new version because many features were stuck in limbo. They realized this had become a meme and that all of their company’s efforts would be a waste if their core was not the one that won, and that the only way to win was to put out a major new release that addressed all common feature requests. They made their case to the CEO and took on the initiative independently to push the new version out even though they were not a manager. The new version came out and the other companies in the space are now the walking dead.

3. Business. One last engineer was the first technical hire at a first-of-its-kind company in fintech that is now valued at near-decacorn status. They were not content to just build features. They took the time to understand the business just as well as the CEO, understand competitors, understand other companies in the space, and even talk to customers to learn what their real problems were. They came to understand what they felt were deficiencies in the business model of a whole class of fintech startups and saw an opportunity to build something new. They founded their own startup that is doing very well.

Good founders look up in all these ways, but looking up does not necessarily mean you will become a founder. Founders have a personality type that requires grit and initiative, and entails an appetite for risk and responsibility. Looking up means you have those personality traits, but not necessarily to the extent needed to become a founder on your own.

Looking up is so important because often the most important factors for success are not technical. Engineers who look up are arguably more impressive than the absolute wizards. Not because they are “better” but because they seem to go on to do the most interesting things.

What’s in a valuation?

To best understand what’s going on with startup valuations, it’s important to have a framework for what valuations even are.

I think Alex Rampell at Andreessen Horowitz has the best description I’ve heard: startup valuations aren’t valuations at all but rather out-of-the-money call options. That is, they are bets that the startup’s valuation will at some point greatly exceed the “valuation” of the round, discounted because it is far away and uncertain. For that reason, I’ll refer to these as “valuations” in quotes for the rest of the piece.

The other side of startup financing is that it’s not just a call option, since derivatives don’t have scarcity (the only limit is the number of people willing to take the other side of the bet) and don’t provide primary capital that the business can use to grow. So unlike derivatives, which are purely zero-sum bets, startup financing has an impact on the future of the business.

Unlike with call options, there is a limited supply of shares that several firms are competing with. So of course, as in all things, the rules of supply and demand apply, and these “valuations” are also prices. Theoretically, there is a perfectly rational price because the terminal value and likelihood of success are objective facts — the variation around pricing come from differing opinions on those two factors.

There are therefore a few evergreen dynamics:

1. If you believe the terminal valuation is higher than before, the “valuation” should go up

2. If you believe the likelihood of achieving the terminal valuation is higher, the “valuation” should go up

3. If more firms are competing for limited supply, the price, or “valuation,” should go up

4. If you think you have better information than other people, or a leading signal, that should inform the price you are willing to accept, especially in a competitive market

5. In an uncertain situation, power-law distributions in terminal values should make you willing to accept a higher price than you normally would if you are thinking about a portfolio in a world with uncertainty

6. If more capital helps the company fuel growth and pull forward, a bigger early round should be rational because it improves the likelihood of attaining terminal value faster

7. If the cost of capital decreases, then implicitly the cost of pulling forward growth decreases, and you should be willing to accept a bigger round

What’s changed and how to react

So what about these huge valuations? While they are certainly not common, they are also no longer rare. This suggests something is different.

Looking at the above dynamics, it becomes clear what has changed. It comes out to massive TAMs, improved signals for established models, and competition.

The first, and obvious one, is massive TAMs that have exceeded everyone’s wildest expectations for decades. The original Uber seed deck imagined a TAM that is less than Uber Eat’s run rate alone. SaaS companies trade publicly for billions of dollars. The older, bigger tech companies are worth literally trillions of dollars. Best in class companies like Snowflake and Cloudflare post metrics that were thought to be impossible. The perceived terminal values of these companies has gone up dramatically, leading many investors to adopt Peter Thiel’s maxim that one cannot overpay for a generational company. And with the technical impact of zero interest rates, the discounted value of those cash flows are up too. Add in markets that are so large that they no longer seem to be winner-take-all and you have many reasons to increase the “valuation.”

The second change is in the quality of signals. It isn’t the wild west anymore in terms of many of the top tech business models. Many of the most popular models, like consumer internet, enterprise SaaS, and UGC marketplace, have spent the last two decades maturing and are now well understood. The metrics that matter are known, the customer bases are well-understood with repeatable rolodexes, and the number of experienced practitioners are large. This lowers risk, and certainly these businesses are easier to understand. For good investors who add value, more signals in a mature space means less reason to care about price since they can just pump money into their machine (at least, if you believe some VCs actually add value).

Let’s take the example of Airbyte (not to pick on them, of course). When it raised its now-famous Series B, Chetan Puttagunta, the Benchmark wunderkind who sits on its board, tweeted a number of non-revenue statistics regarding its open source product. Airbyte is an open core enterprise SaaS company. For someone like Chetan, who has been on the boards of several multi-billion dollar companies that used that same strategy, this seems like a strong signal. He probably believes, very rightly, that as long as the open source project is going in the right direction, he can help them achieve incredible revenue numbers at essentially any point of their trajectory. And if, like Chetan, you are a believer in the idea that software markets are “gigantic” you should be fine thinking that these companies can use cash like a crank and reach stunning valuations. Since primary capital goes towards growth and expansion, if your theory is that you understand how to crank that growth, providing more capital should lead to a higher “valuation” because the likelihood of reaching the terminal value goes up with the increased cash injection.

But there is a darker side to this dynamic too. Funds have more capital than ever, so there is simply more competition. What I am seeing in the market is a willingness to be speculative and allow for degraded signal quality. I have seen this the most with web3 and fintech. Companies with 100x+ revenue multiples have become not rare, as are companies worth $100 million that have not yet built a product. With crypto companies, fees and trading revenue are treated like user engagement numbers. Most notably, I have seen a number of companies with signed contracts that are not yet executedtreated as credible forecasts to justify high revenue multiples on the theory that they reflect low forward revenue multiples. It is important to note that these are real signals, but they are not as good as gold; there should be a discount on these signals that investors aren’t applying.

Competition is forcing investors to jump on these signals, partially because diligence timelines are compressed and partially because the next entry point may be even higher — or not exist at all. This leads investors to view the risk in whole categories as relatively low when with startups the risk is always high. Especially when prices are high, failures in execution can mean companies that are ultimately successful but relatively poor in outcomes for everyone on the cap table. Competition can also lead to ignoring signals too. As interest rates are rising, public markets are possibly going through a period of secular multiple compression, which would affect terminal values. Right now private market “valuations” have not adjusted because competition is too fierce, but if this appears to be non-transient, crossover “valuations” will change.

There is one last dynamic to note: venture capital is a product. What it sells to LPs is a way to deploy cash. So it is possible that investors are deploying cash to fulfill the LP’s side of their business and just foie gras-ing their startups in a way that will be negative for the whole ecosystem in the medium term in order to let themselves raise a new fund. This is not a good long-term strategy for the funds, either, though, since this may not even work for one cycle, let alone two.

The response to this is basically to not compromise on quality and move upmarket. Keith Rabois calls this shifting quality over fighting valuation. There are deals where companies have premium valuations but strong reasons to believe they have technical or market moats and are showing lots of signals. For those companies, the price will remain premium, but the long-run outcome will justify it. On top of that, I expect the rounds to be big, and therefore I expect playing the game to require significant funds, even more than before.

People are asking a lot of questions and opining about whether SPACs are here to stay, whether they are a good thing, and so on. But, assuming SPACs do survive, there has been basically no discussion about what the long-term equilibrium looks like.

A quick process review: IPOs vs SPACs

An IPO is a highly complex, regulated dance. It actually starts more than a year before the IPO itself, usually involving hiring an experienced CFO, formalizing one’s books, and socializing with investors. The company then files with the SEC and hires a bank, called an underwriter, for a very hefty fee. The company then goes on a “roadshow” where they pitch their stock nonstop to large institutions who will buy large blocks of shares just as the shares list on the exchange. Those shares are what retail investors trade on the market. There are many complex dynamics here that I will ignore.1

The SPAC process is rather different.

It starts with an IPO exactly like described above. The difference, though, is that the company that goes public isn’t a real company at all. It is a pool of money that sits in an escrow account for two years that will go out and acquire shares in a private company that, unlike the SPAC, is a real, operating company. The investors in the SPAC don’t know which company it is ahead of time, hence the nickname “blank check company.” The SPAC and target will then go through a reverse merger, bringing the real company public, a process called a de-SPAC, where the owner of the SPAC, called the sponsor, often joins the board of the newly merged company.

SPAC history and the narrative mirage

SPACs were created in the 1990s by GKN Securities as a modern version of a blind-pool offering, but the heyday of SPACs was the Dotcom Boom. They were speculative vehicles with negative adverse selection towards companies without the financials to withstand the IPO process. In other words, during the Dotcom Boom, the “S” in SPAC stood for “shit.” Only the worst companies chose to SPAC. Not surprisingly, SPACs collapsed with the Dotcom Bust. In 2010, only 2 SPACs came to market at all.

The rebirth of SPACs began not with Chamath Palihapitiya but with Bill Gurley, the legendary partner at Benchmark. Companies are rediscovering that going public has benefits too, so interest is increasing. But Gurley noticed that IPO pops are huge, exceeding 20% in the first day of trading.2 IPOs are fundraising mechanisms that bring in real cash for companies, so that means that companies that IPO with a pop are theoretically leaving money on the table — that is, that underwriters gave preferred pricing to their insiders to the company’s detriment. Gurley is today principally a proponent of direct listings, which is an like an IPO but with no bankers.3 He got so upset he even held a conference in San Francisco on direct listings. He has made real progress.

While Gurley was working on the direct listing, two key players were working in parallel on an alternative approach. Diamond Eagles, a highly respected finance team, and Chamath, founder of Social Capital and podcast “bestie,” tried to fill this gap in an unorthodox way: instead of reforming the IPO process, they sidestepped it by reviving the SPAC. Research has shown that financial instruments diffuse through use, which grants acceptance and legitimacy. They showed that in action — they did not just talk about SPACs. They created them. And using their network, they were able to bring highly desirable companies public — Virgin Galactic for Chamath, and Draft Kings for Diamond Eagles. The result is a SPAC boom not seen in 40 years. In 2020, over 50% of IPOs were for SPACs. The bad reputation of SPACs turned out to be somewhat of a narrative mirage. It turned out SPACs had professionalized in the interim. Big companies like Burger King went public via SPAC and reputable firms like Pershing Square and TCG have been backing SPACs for years.

The surge of companies is a response to supply and demand. There used to be over 8000 public companies; now, despite more retail demand than ever, there are fewer than 4000. The main reasons are regulation and private capital. Sarbanes-Oxley increased the cost of being a public company4 in exchange for benefits in reducing fraud, giving public markets a bad reputation compounded by a perception of short-sightedness. At the same time, venture capital funding increased by a factor of 5 since 2010, so companies could stay private longer. This also means there are way more unicorns who could plausibly go public, and indeed in yesteryear would have already gone public years ago. But IPOs are expensive, so a lower-cost approach was natural to evolve to meet public market demand.

Today, SPACs are not garbage. While there are certainly questionable companies that have gone public, especially in the electric car space like Lordstown Motors and Nikola Motors, some of the companies that have SPAC’d are hot companies, including Opendoor, Aspiration, Nextdoor, Hippo, Bird, and others. SPAC share prices have underperformed overall, but some have done quite well.

The fundamental dynamics

There is only one truly fundamental difference between SPACs and IPOs. That difference is an inversion of liquidity on the supply and demand side. The upshot to this is that, in the long run, IPOs and SPACs will select for different types of companies differing primarily by stage.

In the IPO process, recall that companies wishing to go public pursue underwriters to go public. These investment banks act as funnels, managing the process and linking companies to institutional investors. The companies are the demand side and the banks are the supply side. There are far more companies that want to IPO than banks and good bankers, giving significant leverage to the banks. The SPAC process is the reverse. There are lots of SPACs who need a company to take public in a two year window. But there are still not quite so many companies that want to go public. Here, the SPACs are on the demand side, and once again there is much less supply than demand.

There are lots of other things that make the SPAC process different — S-4 filings that are less rigorous than S-1s, forward looking statements, redemption issues that incent celebrity sponsors, differences in diligence standards, complex warrant and PIPE economics, and much more — but the vast majority of these things have to do with the regulatory dynamics and growing pains.5 They are not fundamental to the SPAC market because the SEC could change it overnight. That inversion of supply and demand is the core market chacteristic of SPACs. Some fear this dynamic, saying that it creates inherent negative adverse selection, which the SPAC industry could not survive long-term. This might turn out to be true, but perhaps a more interesting question is: what does a stable equilibrium look like where the SPAC industry does endure?

At a stable equilibrium, a SPAC probably has to be costlier than an IPO or direct listing to incentivize the creation of enough SPACs. They certainly are today. The costs are opaque and not necessarily cheaper than IPOs, especially when accounting for dilution. They also don’t solve the “pop” problem. The average PIPE performance in 2020 was 46%, which was more than double the 2020 IPO pop! To some extent this makes sense. SPACs are riskier and many of them won’t find a target. There are also significant costs in maintaining a management team and board for the SPAC itself, marketing the deal to maintain redemptions, and more. What this does mean is that SPACs can’t be the cheap IPO replacement, so they have to serve another purpose to survive.

There is only one way to maintain an equilibrium with higher SPAC costs, inverted supply and demand, and sufficiently positive selection to make SPACs sustainable. The answer is: selection not on quality but on stage.

Which companies want to go public, are high-quality enough to sustain an industry, will care about which sponsors are on their board, don’t feel ready to go through the gauntlet of the IPO process despite likely lower costs, and can generate great returns? They are crossover companies that want to go public 1–3 years before they are ready to actually crossover.

SPACs would then look a lot like growth equity. They will have higher variance and volatility than IPOs, but may have higher average returns over time. Diligence will continue to improve now that the initial frenzy is over. And just like with growth equity, being able to promise a great board member can help win a deal — so sponsors like ReInvent, founded by Mark Pincus and Reid Hoffman, will probably get better deals than Whoever SPAC Capital Partners IX. In the long run, SPACs aren’t competing with IPOs — they’re competing with private crossover rounds.

This would also require a change in the incentives of SPAC managers, who can still make money even if the de-SPAC’ed stock prices sink. This is the reason people fear negative adverse selection: if sponsors are incentivized to consummate a deal, even if the deal will do poorly, then bad companies will got to SPACs to get a quick deal. For SPACs to avoid this, quality teams need to bring quality companies public and the market needs to not tolerate economics that make SPAC sponsors money regardless of stock performance in time to avoid tarnishing all SPACs.

If this is true, the long run of SPACs is bright, but not an IPO replacement. If this isn’t true, then one of these assumptions about the long-term dynamics is wrong — or there is no stable equilibrium and SPACs will once again be relegated to obscurity and scams. But it seems reasonable to think the market will find a way. After all, we need more public companies.

This is the rise of the full-stack VC.

Full-stack VCs are a currently-rare structure where VCs raise lots of money, invest across every stage, are free to invest in many types of assets like secondaries or tokens, and (usually) employ more staff to help their portfolio companies.

Why do LPs want this? Primarily, it is because venture is the best performing asset class and tech is the best performing sector of public markets.

Why do VCs want this? Good startups are staying private longer, necessitating all-stage investing; much of value accrual is now outside of preferred stock; and deals are more competitive than ever.

The upshot is a significant increase in the complexity of the finance of technology. VCs are responding by becoming full-stack. It is the biggest change in VC since the creation of the field1 and it is going to transform the tech industry along with it.

The full-stack VC model

Venture capital has not innovated as much as the tech industry itself (though the common refrain that venture hasn’t innovated at all is wrong). The simple reason is regulation. Venture capital exists under a very narrow set of SEC exemptions that allow them to avoid the costly regulation that would normally accompany their line of work. The mundane reason is that, until recently, there haven’t been many VCs, and competition is the mother of innovation.

But there is also a deeper reason. The tech industry trades off business model simplicity to enable technology and product innovation. Compared to other sectors of business, the tech industry is relatively simple and sometimes even unsophisticated. As tech more deeply penetrates society, and as companies stay private way longer, this is no longer tenable because the increase in complexity means a need for more types of finance.

The response to all of these things is the full-stack venture capital model. Roughly put, the idea is not just investing across all stages but several asset classes while taking on more operational and financial complexity.

The core concept revolves around the idea that companies now engage in many financial operations beyond primary venture financing, which itself is now occurring much more in the lifetime of a company (requiring investing in all stages to avoid being diluted into oblivion). But companies are now embracing secondary sales and token sales, and they are now starting to look at bonds, factored loans, and more. Because full stack VCs already have relationships with the best startups and understand them well they are also in a strong position to offer such financing.2 It’s also impossible to not mention crypto here. The lesson for tech investors is that they may spot great opportunities to invest in technologies that don’t involve buying preferred stock and they must be able to do so.

Full-stack VCs have a number of different strategies depending on which part of the financial stack they want to attack. Andreessen Horowitz has a wide range of specialized funds like crypto and bio and is particularly focused on buying tokens. Sequoia is focusing primarily on full-lifecycle support and holding its investments basically forever. Iconiq invests heavily in other assets related to tech, like real estate. Most interestingly, General Catalyst is focusing on alternate financing methods, including a credit fund similar to Pipe.3 Over time more funds will compete with them and offer more financial products, like pure debt underwriting or ETFs or maybe even insurance, and invest in more assets directly. By freeing themselves of their previous SEC restrictions, full-stack VCs will also be able to engage in far more complex financing themselves (for example, most VC funds cannot take on leverage).

Full stack VC has a preliminary stage that seems similar but is actually distinct: the all-stage investor. These are funds like Founders Fund, DCVC, Lightspeed, and more who have raised mega-funds or several funds to cover companies in any stage, often doubling down or leading successive rounds in existing portfolio companies or at least maintaining ownership targets across what is now a very long company lifecycle. Some firms, like Benchmark and Homebrew, have fought valiantly against being all-stage despite doing well in order to keep their focus, partially through opportunity vehicles.4

While all full stack VCs are all-stage, not all all-stage VCs are full stack. All-stage VCs are still traditional investors, so they can only invest in primary equity rounds. Full stack is defined by its ability to invest across asset classes. Also in contrast, many all-stage funds are relatively lean compared to full-stack VCs. Plus some tech companies interact with regular PE funds and when they go public or need debt interact almost exclusively with traditional financing companies. In other words, as tech has scaled, its financing universe has remained highly sporadic.

Full-stack VCs are breaking through walls that they have hit against as VC hits its scale limits. Lots of the returns in tech are outside of venture capital, even for portfolio companies. Take Sequoia, which has recently5 had several portfolio companies whose public performance outclassed Sequoia’s private investment,6 like Square, which is specifically mentioned in the memo and returned 41x post-IPO against Sequoia’s 14.5x return upon IPO. For companies like Andreessen Horowitz, who backed Coinbase early on, they would have returned twice as much just buying Bitcoin directly.

The full stack model is more operationally intensive. In addition to higher compliance costs, operating across more asset classes requires more staff to understand and execute those operations.7 There is also competition to offer more support from full-stack VCs because founders now expect more from the best investors, so the ante is up for full-stack VCs to earn a cap table spot.8

Being broad-based has one other interesting implication: scale. It is a direct repudiation of the common wisdom that “venture doesn’t scale.” While many, most notably Sam Lessin, has talked about these advantages, many others come to mind, including: easier international expansion; attracting a wider investment base; investing in other feeder funds, especially for departing GPs; more lobbying clout on behalf of portfolio companies; opening up of direct asset financing opportunities and asset-heavy opportunities; commoditization of back office support; and cheaper financing from a broader asset base, among many more. Scale is good for tech companies and it will probably prove to be good for tech investors too.

The future of VC: where does everyone fit?

To simplify, we can think of VC as splitting among two directions: specialized vs polymaths, and value-add vs “just take my money and run.” These characteristics are as follows:

1. Specialized funds have a focus. Three key methods of specialization are: focusing on a certain stage, focusing on a certain vertical or sector, or focusing on a certain strategy.

2. Polymaths have a broad focus. For the most part, they are cross-stage. Increasingly, they also invest across asset classes and the private-public divide, but not necessarily.

3. Value-add investors bring value other than money.9 Sometimes it is something specific, like solving a certain set of problems through direct work; sometimes it is just connections and advice.

4. Just take my money and run investors know that cash is green. They just give money and get out of the way. This is relative, since even passive VCs still often at least provide introductions and advice.

Below is a graph of where some firms sit with representative classes for many types of firms. The top right corner is where full-stack investors live, though some great all-stage investors live there too.

The obvious question is: what room is there for new entrants? VC fundraising is at an all time high, but since 2018 new fund formation has fallen off a cliff, below even 2010. Full stack VCs keep upping the ante and can keep funding their best companies over and over, and specialist funds keep getting better and better. There is increasingly less white space. But new funds are a critical part of the venture ecosystem, so it is in the best interest of full-stack funds to nurture this part of the ecosystem. Expect to see much more investment in VC funds by full-stack VCs, perhaps even supplanting some of the best fund-of-funds.

Full-stack VC investing opens up two deep, interesting questions.

First, full-stack VCs are the first to take “investing in tech” as a universal concept seriously. Typically, LPs think about investing as occupying a certain risk-reward profile. As a result, they allocate their funds based on the expected blended return. Full-stack VC blows up this model because it is not purely “venture” as previously defined and therefore theoretically has a different risk profile; truthfully, we should call these new beasts full-stack tech investors. If tech is now an asset class in itself, LPs will eventually start asking questions about which sleeve the full-stack VCs should draw from. It also will call into question the expertise that full-stack VCs need because they will begin engaging in complex transactions that involve significant execution risk. Full-stack VCs will probably start asking themselves questions about where they belong, too, like whether they should still support the NVCA and whether they will professionalize. While the conversation right now is mostly about the advantages of being broad-based asset managers, there are disadvantages too: it could make VCs too comfortable and reduce their ultimate rates of return, which is what has made VC so attractive in the first place.10

Second, as full-stack VCs get more complicated, tech might as well. Tech financing is simple: remember, Silicon Valley has thus far traded business innovation for product and technology innovation, but outside of tech there is a lot of complexity, especially in terms of operations, finance, and legal. If software truly eats the world, you are what you eat, which will ultimately mean more complexity in what tech companies and their investors do. Software is no longer a cottage industry, and so it may evolve beyond paleolithic finance because venture investors will finally be in a position to offer it just as the industry is becoming mature enough to digest it. By adding additional degrees of freedom to business success, there are opportunities for tech companies to emerge that do not need technical innovation to succeed, which would change the complexion of all of tech to look more like other industries.

Whichever way it goes, no one can say venture is a sleepy industry today; it is just now birthing a whole new flavor of offerings. Full-stack VCs are really full-stack investors in tech as a whole, and their mere existence is primarily a testament to how much it has scaled. Tech is now so big it won’t even fit the same finance.

Some people are noticing there is so much money, and VCs are complaining about there being “too much capital.” Deals are closing super fast, pricing discipline is going away, and diligence is decreasing as investors all pile on.

At the same time, when you talk to founders offline (as opposed to Twitter where everyone wants to signal that things are easy) they will say it is quite difficult to get started. More “early stage” VCs are no longer doing pre-seed while more angels want to see traction.

So where is the disconnect? Why are many early stage founders experiencing difficulty while VCs are amazed at how much money is around?

Something interesting is happening. To find the answer I looked at the numbers in the NVCA 2020 Venture Monitor.

There are indeed more dollars than ever, and the oversubscribed rounds are getting more oversubscribed. But the actual number of rounds are going down.

As you can see, there were fewer deals in 2020 than in 2015. In fact, the average deal size was 89% higher.

What’s driving this massive increase is the rise of mega-deals, which have tripled since 2015 and which now dominate the headlines (and therefore perceptions). More notably, mega deals were only 29.7% of all venture funding in 2015; today, it’s 47.9%. If you look at the data room, you can see that early stage VC has decreased as a percent of deals, but only modestly. There was a clear hit at the beginning of COVID-19, but early stage is only a little below where it was in the deal mix, and there is still a monotonically decreasing relationship between deal stage and deal count.

What this means is that there is more money than ever but it’s chasing fewer deals, mostly in growth and late stages. So valuations are going up in late stage but not in early stage deals; in fact, early stage valuations dipped this year even though early stage deals exceeding $25 million hit 65.7% of all deal value, an all-time high, thanks to increasing consensus on the top deals and the entry of players with lots of dollars, like Tiger, Altimeter, and SoftBank. The average, in other words, is hiding yawning inequality.

There is also something that isn’t captured in the numbers. Companies are jumping valuations very quickly. Companies like Ramp are raising rounds within months of each other and companies like Hopin are hitting unicorn valuations in just a few years, all while Tiger is doing a deal a day. VCs, who are seeing these dynamics every day, feel this as part of the broader market while founders only feel it for their startup, where individually it may be getting harder and harder. On the founder anecdote side, many of the top funds still have websites that say they “invest at the earliest stages” but now hardly ever invest before the Series A. This further affects founder perceptions of the market being relatively dried up.

So that’s where the disconnect comes from. Both sides are seeing the market correctly, just from different angles: VCs are repeat players so they see it longitudinally while founders aren’t so they only see cross sectionally. If you are a VC, you have never had more competition for the deals you want to get into. But if you’re a founder, especially in early stage, you just see a market that still hasn’t rebounded.

The issue with these businesses is a coordination problem. To get the businesses involved, you need to show them that customers want it, and to get customers to demand it you need to have businesses already onboard. To make things worse, the buyer is often an administrator whose incentives aren’t aligned with either side of the product. Often, B2B2C is an attempt to get the market size of consumer with the predictability of enterprise, but agency problems mean that these businesses often fail to achieve either. The core solution to this problem is to ignore the B2B2C aspect of it and pick a side. Then, you dominate that side and become what Ben Thompson calls an “aggregator” because you own a set of customers.1 The B2B2C part of the business then comes easily. You can just build whatever additional functionality you want since, hey, you already have one side of the market, and owning that side of the market gives you distribution. Chicken, meet egg.

In order for this to work, your solution has to have some value prop other than the integration. What’s that mean? Instead of one B2B2C product with two features for each side of the market, you’re actually selling two products to two markets, a B2C and B2B product, that happen to include one interaction feature that connects them. For that to work, your early engineering and product vision has to be around something that provides independent value for one side only. One of my favorites is Withings. They started as a pure B2C company, making products like smart scales that customers wanted independent of any integration with their healthcare providers. Now, they offer robust features to share your clinical-grade data with your healthcare provider and are well on their way to being a true wellness B2B2C play.

There’s an alternative: verticalization. Instead of selling your better solution to a business, just build your own. In the education market, that would mean building a school instead of selling something, like presentation software, to educators at an existing school and trickling it down to students.

My poster child for verticalization is OneMedical.2 Much ink has been spilled on how painful it is to create a new EHR system.

But what if you really did build an awesome EHR system? If you were incredibly confident that this would help you compete, perhaps you wouldn’t want to sell your EHR to other providers but would just start your own clinic using your software, outcompeting the simple doctors who don’t want to use your product. Interestingly, this is how OneMedical’s parent company, 1Life, describes itself:

1Life’s heart is an EHR and charting system designed and built alongside clinicians to be the easiest way to record and read patient care information. But that’s just the beginning. The foundation of our EHR is a modern tech stack, built on Rails, implemented with technologies like GraphQL and ElasticSearch, all deployed on top of AWS infrastructure. In addition to the EHR, 1Life includes our scheduling, administrative, tasking, and messaging platforms, as well as our our patient web portal and native iOS and Android apps.

In other words, OneMedical’s killer piece isn’t just its awesome doctors, its partnerships, or any of the normal things medical care companies (including OneMedical) talk about. It’s the backend software that enables those things — the kind of tech that most founders would try to sell to clinics, but which OneMedical instead uses to directly compete with those same would-be customers.

Another verticalization success is actually Uber. The original pitch mostly describes the technological failings of an industry. Here is how Garrett Camp originally described it:

There is no notion in the pitch deck that then-UberCab was ever considering being a vendor for the aging cab monopolies,3 even though the problem slide easily could have been for a software vendor pitch. Lots of companies, in fact, would have gone on to discuss how many cab companies there were in the US with a GTM slide about a pilot with a cab company, say Yellow Cab Corporation. Instead, UberCab recognized these as weaknesses with the existing regime and was so confident that its solutions would be definitive that they created their own competitor and won.

With verticalization, there is no confusion as to who the actual customer is. With B2B2C there often is. The B2B and B2C products will have competing product choices that depend on which side of the market is more elastic and, often, inertia. Verticalization sidesteps these issues by eliminating the inherent tension of multi-sided customers: there are only users and vendors instead of two sets of customers.4

Why are founders seemingly averse to this approach?

1. Historically, SaaS was viewed as the low-risk, low-return alley of the tech industry with the big dollars coming from consumer Internet. Today, enterprise is having its moment, which makes the risk-adjusted opportunity more appealing.

2. Verticalization means giving up software margins of 80%+ for the margins of whatever business you’re verticalizing. On a multiples basis, this may make the business less attractive. But since your gross increases the total business value may actually go up.

3. Since so many founders are software engineers, this means building fundamentally different skills. If you’re building a verticalized salad restaurant like Sweetgreen instead of just selling supply chain software, most of what you need to do isn’t what you already know.

4. As David Sacks has noted, tech-enabled businesses are somewhat new, but investors are getting more discerning. Founders are probably pushed in this direction at least partially due to investor preferences.

5. On the downside, there are many improvements that are significant enough that existing businesses will pay for them but not dramatic enough to get customers to shift their behavior. Ex-ante, it can be difficult to know which one you have.

6. Ideally, the tech generates new supply. This might not always be true, and is difficult to gauge ex-ante. For software this is almost never a problem, since figuring out whether the problem can be solved is just an engineering challenge.

Of course, there is one other important problem: verticalization can be more expensive and risky. Companies like Atrium and Altschool tried to verticalize and failed. Why? Verticalization turns your company into a services business, and you need to be good at the actual service. That sometimes requires significant money to get to an MVP and service your early customers (Atrium and Altschool both raised around $100m before failing). Even if your software creates a better experience or a better cost structure, the value still needs to be there. By not being a good enough law firm, or having the wrong school model, verticalization meant failure.

This is a particularly promising time for virtual verticalization. There are a number of businesses that would normally be interesting to verticalize, but the current crisis has opened opportunities to do so without an upfront brick-and-mortar investment, lowering the upfront development cost.

Two interesting examples, though there are others. One is telemedicine-enabled verticals, like psychiatrists. American telemedicine is primarily a prescription tool. Sure, there’s Teladoc, but this has been pretty limited. One of the biggest reasons is that telehealth is billed differently than in-person care, but now this is changing — the government issued an emergency order for COVID-19 and now the government says this may be here to stay. Another is schooling. People like Austen Allred have mused that there’s a market for homeschooling. The rub is a three-way wedge: most parents are ill-equipped to be good teachers, tutors are too expensive for most families, and MOOCs aren’t personalized enough for kids. Well, guess what kids have gotten quite used to over the past few months? Online schooling. Teachers are notoriously underpaid relative to the value they provide, but they would be downright cheap by the standards of most service suppliers. On top of that, education is the ultimate B2B2C failure since the administrator-buyers are some of the least-aligned purchasers in existence.

Essentially, founder reticence comes down to the pithy phrase about picks and shovels during the gold rush. The meme is right, but if you’re the inventor of the pick you are sometimes better off being the first good prospector.

The core idea is simple. Instead of a single fundraise establishing a committed capital base, venture capitalists can garner quarterly commitments, giving them fresh capital each quarter to invest. This is a radical departure from how traditional VC firms work.

The biggest rolling funds are actually quite sizeable. Sahil Lavingia, founder of Gumroad, expects to deploy $8 million a year — which is the size of a proper pre-seed fund!

Here is how AngelList itself describes its creation:

Similar to how SAFEs brought high-resolution fundraising to startups, Rolling Venture Funds brings high-resolution fundraising to venture funds. With a Rolling Venture Fund, fund managers can now accept new capital in the form of auto-renewing quarterly commitments.

These are big words! SAFEs were arguably the most important structural innovation in venture in the 2010s. Will rolling funds similarly live up to that comparison and be the transformative structure for venture in the 2020s? That’s the 130 billion dollar question.

So, a week after the anniversary of their launch, let’s review: how should we think about rolling funds?

What are rolling funds?

Rolling funds are based on a single premise: instead of raising a whole fund at once, fund managers can raise the fund one quarter at a time. This is where the “rolling” part comes in: the opportunity to invest comes once a quarter instead of being fixed in time with dedicated close dates and painful roadshows.

Technically, venture capitalist firms raise several venture capital funds — for example, Sequoia Capital will raise Sequoia Fund I, Sequoia Fund II, etc, and formally it is this legal entity that invests.2 The partners at Sequoia spent months (for less famous venture capitalists it can take years) talking to big institutions and convincing them to commit money to the fund,3 and they agree to manage the fund for a contractually defined timeframe. This process is extremely painful and a huge barrier to entry — to be a VC, you might need to be able to sustain yourself without receiving any income for years while you fundraise. And since you have to raise it all at once, if you don’t succeed at raising a huge sum you will not be able to successfully create your fund at all.

The traditional approach introduces three problems:

1. The fundraising barrier to starting a fund is extremely high

2. Funds are constrained by the legal mechanisms that enable this particular form

3. There is a lot of backend complexity to running a fund

Rolling funds solve each one of these problems, creating not one but three value propositions for fund managers:

1. A new capital structure that allows for much faster and flexible fundraising

2. A legal structure to accommodate this kind of fundraising

3. Backend rails to let fund managers invest without building a back office

Rolling funds are entirely managed through AngelList. Fund managers define their closing conditions, like the minimum investment size. Investors can also pre-commit to funding several quarters at once, giving star rolling fund managers something resembling a permanent capital base.

The way this is legally handled is not by a rolling close but a vast proliferation of entities. In other words, each quarter, AngelList creates a new limited partnership to manage that quarter’s commitments. This is necessary because if you invested in Lambda School in Quarter t, you would not want the investors who didn’t take the risk early on to be able to invest in Quarter t+8 and get a share of the returns.4

As you can tell, this creates meaningful administration complexity on top of the existing madness of running a fund. To deal with this, AngelList leveraged its experience with syndicates and funds to create a back office for rolling fund managers. You don’t have to leave their platform: they’ll create the legal entities, hold the committed capital, wire the funds for you, and more. They even have staff to, for example, provide common side letters. This last piece is rarely discussed but it’s essential — if the point of rolling funds is to decrease the overhead of running a fund, reducing the administration to zero through operational experience and technology is a critical piece.5 AngelList already had most of this infrastructure for its fund product and simply repurposed it.

Due to the low capital commitment and flexible investing structure, rolling funds are a sort of scout fund in reverse: a diverse set of investors who can provide support and deal flow to the whole portfolio because they would like to invest but probably couldn’t have met the bite size on their own. Traditionally, scout funds solved this problem by having a big, traditional fund with lots of commitments and contracting with relevant operators, promising some of the returns. Rolling funds also give operators the chance to play under a fund’s banner, but they solve the problem by allowing these same people to pool their money with checks that would be too small to be interesting for a big fundraise — and arguably, this is a little better since it makes the scouts have skin in the game.

By having commitments on a quarterly basis, rolling funds introduce an unprecedented amount of flexibility to VC. Because each fund closes quarterly, new fund managers can start small and build momentum. With the old fund structure this would be impossible because they would have to deploy the old fund, and if they started to see success they wouldn’t be able to capitalize on it quickly because of the cadence of a traditional fundraise. It also means that fund managers could have, for example, seasonal funds (say, a rolling fund manager who only invests in the summers) or take a break and regroup after a rough year or important life event, like a new child or a sick parent.

AngelList is perhaps the perfect firm to have invented rolling funds because they also commoditized the syndicate.6 In fact, you can think of rolling funds as an obvious expansion of syndicates: instead of people banding together over the internet to invest in a single startup, they now band together to invest in a single fund.

Who will this affect most?

Let’s talk about which slices of VC will be most dramatically affected by rolling funds.

The first is micro-VC, which is defined as early-stage VC funds with less than $50m. This is probably the most obvious. The traditional fund route involves a high fixed cost regardless of the dollar amount raised, both fiscal (plane rides add up) and psychic (smaller checks, same boot leather). In exchange for all that pain, you get what, $10 million after a year plus of fundraising? Plus a business structure designed for funds with much larger capital bases? No, thank you. With a rolling fund you can get just as much money, albeit with less certainty for the life of the fund. It makes way more sense at that dollar amount to just take the money as you get it and invest. For example, Cindy Bi, who is a super angel with a strong track record, has sung the praises of rolling funds and the massive increase in speed that they gave her fundraise. That also allows fund managers to establish a track record during the fundraising process, which can grease the wheels for larger, later closes.

The next big group to be transformed is operators. I wrote earlier this year about the rise of operator investors — funnily enough, just a week and a half before the announcement of rolling funds. Founders increasingly want other operators to be their investors. The problem? Operators usually don’t have enough liquidity to invest on their own, and they don’t have the time to raise a proper fund (nor could they; being a current operator is part of their appeal). Because of the flexibility and backend tools of rolling funds, operators don’t have to have the same time commitment they would need if they were running a normal fund too, which means that more operators will have the time to run rolling funds. That said, you already saw bigger founders like Rahul Vohra raising capital before there were rolling funds, so arguably the biggest barrier was the rails AngelList already provided.

And lastly, you have scout funds, which rolling funds will largely replace. Rolling funds aren’t in obvious conflict with the existence of scout funds since, ostensibly, the main function they serve is creating a deal funnel to larger VCs. Maybe so for the funds, but not so for the scouts themselves: their problem is lack of access to capital. Scout funds solve this by giving scouts enough money to invest, but the scale is quite small — 5–10 checks that are tiny and meant to be written at the earliest stages. If the top 300 scouts can raise small rolling funds that give them 2–5x the capital, it’s worth it for the scouts to jump ship. Scouts all want relationships with the top scout funds anyways because their ability to raise next quarter’s rolling fund depends on their ability to say they coinvested with a Tier 1, so scout funds can just keep their relationship with the top rolling fund managers by investing in their funds and remove a lot of their own overhead.7 Scout funds might just be better off becoming rolling fund-of-funds.

Rolling funds create incentive misalignment with follow-on rounds

Rolling funds are an amazing innovation that, as I’ve argued above, will produce a lot of good. That said, rolling funds come with a downside: a longer-term misalignment on follow-on rounds. This will create tensions between rolling GPs and their LPs and affect their ultimate returns.

Typically, the life of a fund is 10 years because it takes that long for the best companies of the fund to mature (which, as Scott Kupor points out in his book, is longer than the typical marriage!).8 During that time, investors need to measure and meaningfully compare the returns to other assets. What really matters is the cash-on-cash returns.9 In other words, when all’s said and done, how much actual money did your LPs make?

How do you get good cash returns? Once you move past the “invest in good companies” part, it’s by doubling down on your winners.10 You need to make sure as many of your dollars as possible are in your winning investments becaise the power law dynamic means that your top portfolio companies will outperform the rest of the portfolio, even at a later stage.11 You can get good returns without doing this, but it’s tough to get great returns without it, and it’s impossible to do so at dollar amounts that matter.12

The core mechanic that creates this problem is that rolling funds exacerbate the dichotomy between fund and firm. Traditional funds have a committed capital base that is full of institutional LPs who are investing lots of money over a long time horizon across several funds run by the same firm. When you have a long-term relationship with the same firm, it is not so problematic to concentrate on a single successful portfolio company, even if it is across funds. In other words, the fund/firm dichotomy matters less. With rolling funds, there is a new fund each quarter, each with a potentially widely different investor base, so that dynamic matters a lot.

Think about it from the perspective of the dollars. When each quarter has a different set of LPs, you want to maximize the performance of each individual dollar in order to incentivize new investors to participate each quarter, which means investments as early as possible to get a great headline that drives subscriptions. But if you are thinking about it from the perspective of a pool of dollars instead, the performance of each individual dollar matters less; what matters is getting as much of the pool as possible into your winning investment.

This can somewhat be mitigated by increasing the subscription term. Instead of quarterly, you may eventually see rolling funds raising for a year at a time. But once you’re talking about 2 year subscription periods you’re raising a fund that isn’t really “rolling” anymore — it’s just a regular, 2–3 years to deploy fund that didn’t have any follow-on reserves. LPs can also precommit to several quarters of funding, but since each quarter is a new LP there’s no way to know how much of it will be going towards follow-on investing. And since there’s total flexibility on the part of LPs, GPs don’t have that much predictability as to the follow-on capital they’ll have in 5–7 years when it’s clear who the big winners are.

There are other issues with the rolling fund dynamic that make it difficult to double down. You can’t guarantee the timing of your fundraise, for one, and you can’t guarantee that your LPs will be around to invest in future rounds. What if you have an amazing investment that doesn’t raise again for a few years so your LPs abandon you? You could also try to raise an SPV, but rolling funds are inherently less private, so you can’t put out a private note without risking jeopardizing the sanctity of the funding round itself, or prioritizing certain LPs, which would violate one’s fiduciary duties. Even superstar rolling fund managers will encounter this problem because even though they will know they can fundraise indefinitely they will not necessarily know the size of the fund each quarter.

We spent a lot of time in previous sections talking about the minimum check size, but in later, competitive rounds the main limitation is actually a maximum check size — the allocation. Good firms not only take their pro rata but are able to get over-allocations at the expense of other investors. With these issues it will be much harder to secure that.

A promising legacy

Venture capitalists often say they won’t fund “lifestyle businesses” that have no real way to scale to large returns — rolling funds actually enable, for the first time, lifestyle funds. They don’t need huge winners to be successful for all involved, and by allowing for an increase in the diversity of fund managers they mean that more lifestyle businesses can be funded too, improving the ecosystem overall.

But for rolling fund managers who want to make it big, they will need to secure follow on capital. It’s basically impossible to become even a hundred millionaire by just cutting a seed check in a unicorn but it’s quite probable if you keep writing follow on checks.13 Due to the follow-on problem, the rolling fund managers who truly want to scale will eventually have to abandon their rolling funds for a more traditional route.

What this means is that the biggest legacy of rolling funds might be as a new, decentralized talent funnel into the biggest funds. We have already seen blogging and tweeting as legitimate means to enter VC by proving that you can pick good startups. Rolling funds take that a step further by having a costlier signal. You actually put capital at risk and have to show that you can actually get into rounds yourself, not just talk about who you might like, without already having to be rich enough to angel invest or needing fancy credentials. There is no theoretically stronger signal, so rolling funds might enable a new class of fund managers to come from anywhere.

That would be an amazing legacy in itself.