Be a winner, or join one?

One shot, one opportunity.

Kansas State?

Your whole future depends on what you are going to do in the next hour. Nothing less. You have got to strike for your freedom, and strike now, for the moment may not come again.

– Carmen Maria Machado, In the Dream House

—

For better or for worse, all the famous thought experiments are real now:

• If you could plug yourself into a machine that offered infinite entertainment and pleasure, would you do it? Is this machine good? If the world around us is fake, can our feelings about living in it still be real?

• If an English speaker is put in a room, given a message in Chinese, and told to respond to it using a precise flowchart that tells them how to do it, does that person understand Chinese? If the instructions are so comprehensive that they’re able to respond to any Chinese message as well as a native Chinese speaker could, do they then understand Chinese? What if the person uses a second set of internal algorithms to navigate the flowchart? What if the flowchart is probabilistic, and the answers appear not only correct, but novel and creative?

• If a chatbot is trained to act exactly like a friend—you can text it; it frequently texts you back; if you text it at night, it usually doesn’t respond until the morning; it goes offline some days; it often agrees with you, but sometimes it doesn’t; sometimes it is grumpy; sometimes it texts you first, and asks you for favors; it remembers what you told it, but imperfectly so; it cannot be manipulated by its maker; it cannot be deleted, but you can choose to stop texting it, or, if you upset it, it can choose to stop texting you—is it wrong to be friends with the chatbot? Or do we just rationalize it as wrong, because it feels like it should be?

• If someone is struck by lightning and dies at the exact moment that another lightning bolt strikes a tree and, in an improbable coincidence, reconfigures its atoms to be identical to those in the person’s body, does that second being “remember” anything? Have they had real experiences? If you talk to them, who are you talking to?

• Can you solve the mysteries of God?

Anyway, here’s another, less theological one:

Imagine that you are a top college basketball prospect. College coaches are cold calling you; NIL collectives are pooling millions of dollars to sponsor you; Texas Tech is offering you millions more, plus $250 in bonus bets on FanDuel. You can start at any school you want.

You narrow down your choice down to two options. One is a blue blood—they’ve won multiple titles; other top recruits have already committed; they would become national favorites if you signed. The other team is an up-and-coming program that’s never even made a Final Four, and you’d be the biggest star that they’ve ever signed. Do you join the school with a championship legacy, or do you try to build your own? Do you add your name to the first school’s long list of All-American alumni, or do you become the second school’s singular star?

Which is the right choice? Which is the honorable one?¹ Do you join a winner, or try to be one on your own?²

—

Once again, AI brings the experiment to life. The AI supernova has created bigger and more concentrated winners than ever. The last decade of technology investing has been overwhelmingly dominated by three AI companies. Four companies—Anthropic, OpenAI, SpaceX, and Cerebras—were responsible for propelling 29 venture capitalists onto Forbes’ list of Silicon Valley’s 100 best investors.³ “Smaller” startups are also going straight up, and sometimes become industry defaults in a matter of months. In Silicon Valley, there are rides on rocket ships everywhere.

But, simultaneously—it has never been more tempting to try to build the thing on your own.⁴ You can incorporate a company in the first hour; you can build a product in the second hour; you can conjure a workforce of robots to manage all of your administrivia after that. One of the dizzying facts about the internet is that incomprehensible power and unfathomable wealth has always been one long computer cheat code away; now, when you can build stuff online with a few sentences, that cheat code feels tantalizingly close.

So what do you do? Join a winner, or try to be one on your own?

—

This is not, of course, an entirely new question. This has always been a choice for people who work in tech: Join a big company, or start a tiny one; methodically and ruthlessly scale a popular thing, or frantically build a new thing; optimize the edges of a household name, or live on the lonely frontier of invention; get paid in cash, or get paid in lottery scratchers; take a supporting role, or audition to be the lead. Each side had a consistent character: One was big, stable, secure, slightly boring, and becoming state; the other was scrappy, frenzied, volatile, often unnerving, and trying to be cool. There were many tradeoffs between one and the other, but really, it was mostly just one tradeoff, phrased in a dozen different ways.

Now, everything is helter-skelter. The biggest companies are the coolest ones; the trillion-dollar behemoths are the most disorganized; and the billion-user platforms are the most disruptive. The labs and booming startups are white-hot centers of invention and innovation and the escape pods from the permanent underclass, while the smaller companies are the derivative wrappers. In a lot of ways, then, the equation should lean almost entirely to one side.⁵

But then, there is that last point: Does it count, if you join the team that could just as easily win without you? Look at the sea of opportunity out there, they say; “it’s a waste to ignore how much the world has changed.” Are you really going to squander it by working on someone else’s idea, doing what someone else tells you to do? If you came to Silicon Valley to invent something, will there ever be a better time? If you came because of blind ambition, shouldn’t you be chasing it? You only get one shot, etc., etc.

Or is that feeling just the zeitgeist—or, like, you know, venture capitalists—exerting its social pressure? Because courage could cut both ways, and there are two ways to miss the moment: By not dreaming big enough, or by caring more about owning work that may well not matter than contributing to work that almost certainly will.

—

Or, here is another thought. The epigraph at the beginning of this post is a great line from a good book, about a circumstance far more grave and urgent than any of the comfortable trivia we talk about here. And that is worth remembering. These are real questions, yes, but no matter how much weight and urgency the machinery of this industry⁶ lends to itself, it is, in many ways, just a different sort of experience machine—an all-consuming ecology, where it is easy to lose track of what’s real and what’s artificially exaggerated.

1. The answer, it seems, is you’re damned if you do and damned if you don’t. On one hand, the definitive measuring stick for most athletes is how many championships they won. “Jordan has six titles and LeBron has four,” people will say, forever. On the other hand, when players shamelessly chase titles—LeBron building a superteam in Miami; Kevin Durant joining the Warriors after being beaten by the Warriors—people accuse them of being pathetic.

2. Credit where credit’s due, to the original philosopher to ask this question.

3. You can accuse Forbes of making bad lists, but you can’t accuse them of making meaningless lists.

4. Though, as we’ve talked about before, there has perhaps never been a worse time to do it:

The machines aren’t static. Across vendors and release versions, models could develop their own habits. Just as Sonnet 3.5 might be better off unencumbered by us and our feeble reasoning, Opus 4.5 could be better off unencumbered by Sonnet 3.5. And if that’s true, it seems naive to assume that some future AI won’t want to undo the mess Opus 4.5 made. There is no fixed definition of tech debt—it is simply code that the current engineers would prefer to be written differently.

And so it continues. If you recently built an app with Opus 4-dot-whatever, were you tempted to start over with Fable? Do you think you’ll try again with GPT 5.6?

5. But maybe not! You could make this mechanical math problem: What is the salary at the big company; what is the equity worth now; what’s a rough distribution of possible outcomes and probabilities; how liquid is the equity; if you started or joined a tiny company, how likely is its success; over how much time; and what might it be worth? That last number could be very big! But it’s extremely hard to estimate. And anyway, the question of “what do I want to do?” is probably answered by the number that you put in that box.

6. Which includes things like histrionic blog posts.