Agency
Back in 2023, Sam Altman floated an odd prediction: that one person would soon build a billion-dollar company. It sounded like a stunt at the time. Two years later it reads more like a roadmap — and the reason isn’t only the models. It’s the people who’ve learned to point them.
A new species of company has appeared. Tiny, a little feral, improbably profitable. Hundreds of millions in revenue with no sales org, no marketing department, no HR, sometimes not a single specialized engineer on the payroll. A handful of people quietly producing the output of a few hundred, with machines closing the gap.
For a decade we braced for automation to take our jobs. What happened instead is stranger: AI didn’t replace human drive, it multiplied it. And in multiplying it, it moved the line that decides who wins. That line used to be schooling, or pedigree, or some hard-won specialty. Now it’s agency — the plain willingness to act before anyone hands you permission.
Agency
It’s a slippery word, so let me pin it down. My job is building infrastructure for AI agents: programs that read their environment and act toward a goal. One drafts legal strategy from a stack of filings; another watches the market and moves money. We call them agents, and I’ve come to mistrust the label, because agency is the one thing we deliberately keep from them. We give them capability, obedience, and predictability instead. The AI products people actually pay for are the ones that wait to be told what to do — the moment a lab shipped a model that acted on its own, users recoiled.
People with real agency don’t wait. The trait is unruly by nature: you move without sign-off, without instructions, without anyone’s blessing. It’s the quiet conviction that you can simply do things — poke the world and watch something fall out the other side. It’s an investor with no research credentials starting the most important AI lab alive; a payments founder willing a private space industry into being. It runs on stubbornness, improvisation, gut, and a useful streak of irrationality.
You don’t need to work in tech to have it. But that’s where most of these people end up, because they gravitate toward places with little structure and enormous leverage — startups.
What used to hold them back
Ambitious people have always been around. The constraint was bandwidth. The world is intricate, and doing anything real eventually demands a specialty, which takes years to earn.
My own first attempt at building a product ate the better part of a year, most of it spent simply learning enough to ship something crude — and crude is generous. Real fluency takes far longer. That’s why we lean on credentials over results: a specialty is a private moat, and the sensible play has long been to pick a lane, do the assigned work, and inch up the ladder. There’s a reason tuition keeps outrunning inflation. People are paying for the moat.
Generalists never had it easy. Until now.
The phase shift
AI has hollowed out the premium on specialization, because for an enormous range of tasks, the result of years of practice now costs twenty dollars a month. What used to take me months takes a week.
The pushback is that AI is sloppy and probabilistic, that you still need an expert to know when to trust it. That part is fair — but the conclusion isn’t. Specialization didn’t become worthless; it became uneven. I expect the world to split in two.
Where a mistake can kill and the model can’t promise near-omniscience — defense, medicine, spaceflight, biology, frontier AI itself — we’ll keep insisting on a human who’s accountable, and regulation will hold that line. Sometimes we just need a name to put on the failure.
Everywhere else — anywhere “run it again” is a fine answer — the dam breaks. Marketing, financial modeling, design, teaching, analysis: all of it fills with people who lack the credential and simply have the agency. The machines will keep slipping, but they get better at a frankly absurd rate, and every year postpones the moment a generalist has to call in a specialist.
The winning move changes with it. It stops being about mastering the details and starts being about seeing the whole board. Less knowing how to patch the system, more knowing that it needs patching. More architecture, less wiring. Precisely the terrain where generalists win.
So the walls between jobs are thinning. I’ve watched product managers build financial models, designers run ad campaigns, restaurant owners ship their own pricing tools, farmers wire up crop trackers. The capacity was always in them. What changed is that it no longer costs years to express it.
Carry that to its limit and you get one person running a whole company. The share of solo founders has nearly doubled in a few years. Henri Shi, who spent a decade growing super.com into a $150M business, now keeps a public leaderboard tracking the race to Altman’s one-person billion-dollar company — and clocks an average of $2.8M in revenue per head, the very number Apple posts. Midjourney: forty people, half a billion in revenue. None of this is a fluke. It’s a trailer.
Credentialism is coming apart. The advantage is no longer knowing one thing extraordinarily well. It’s the bias toward making the thing happen.
A new world
My entire picture of the world has compressed into one bit: agency, or no agency.
It won’t be a clean or gentle transition. Institutions built on credentials will not bow out quietly. Middle managers will guard their headcount, because we still mistake more people for more importance. Schools will lag for years. Only competition will force the issue.
And structure has its uses. A one-person company is a single botched tax filing from disaster, with no colleague to break the fall. Even so, these operators will be brutal competitors that the incumbents can’t wave away.
The encouraging part is that agency lives in the head, and what lives in the head can be picked up. It’s Morpheus asking Neo whether that’s really air he’s breathing. The boundaries we treated as fixed — degrees, titles, years on the job — were never the walls we took them for.
Like Neo, the hard part was never the jump. It’s believing you’re free to make it.