John Collison said something recently that has been rattling around my head. He said the two categories of people he would bet on for the next decade are high-agency people and double majors. The high-agency people, he said, are the ones with that pep in their step who want to make things better. They have been so much more empowered thanks to AI.
The framing is right, but the mechanism is more interesting than the headline. AI did not create high agency. It multiplied it. The people who already had it just got handed an enormous lever. The people who did not, are now competing against it.
Something specific changed
Big teams, long meetings, six-week briefs were the speed of business because that was the speed of everyone. Then Cursor, Claude, and the agent stack arrived. The slow-org moat went with them.
Pieter Levels, who built PhotoAI alone, hit $100K MRR by September 2024. One person, no team, no funding. There is no version of that timeline that exists in 2019. The infrastructure to ship a real SaaS product as a single human in your bedroom did not exist. By 2024 it did. By 2026 it is the new normal.
Klarna announced in February 2024 that its AI assistant was doing the work of seven hundred customer service agents and saving $40 million a year. The math broke. Most orgs are still budgeting against the old equation: people equal output, headcount equals capability. The math broke because that equation no longer holds.
High agency is three things
Not founder mythology. Not a personality trait. Three things.
Sees the gap. Decides without permission. Ships.
The loop runs in days, not quarters. It looks like this: you spot the gap on Monday. (Parents in N8 cannot find pram-friendly pubs in London.) You build the thing by Thursday. (A voice agent that calls the venues.) You ship by Friday. (Buggy Smart, 1,180 venues, live map.) That is one of mine. The point is not the project. The point is the cadence.
The thing that changed in the last eighteen months is that the cadence is now possible for one person.
What high agency is not
Hustle. Hours are not the unit. Decisions are.
Heroics. The drama is the tell that the system is broken.
Lone-wolf. High agency talks to more people, not fewer. Just not in steercos.
Founder worship. It is a habit anyone can run, at any level, in any org.
The trap most companies are walking into
MIT's 2025 GenAI Divide report found that ninety-five percent of enterprise AI pilots deliver no measurable business impact. Despite thirty to forty billion dollars in spend. The bottleneck was never the tool. Seats bought, dashboards lit, output unchanged.
Tools without agency ship nothing. This is the trap. You can give every employee Claude and Cursor and a Copilot seat. If they still need three approvals to ship a fix that takes an afternoon, you have just bought a faster typewriter for a queue that does not move.
Yes, but
A smart sceptic raises four objections. They are all real.
One: "Without taste you just ship garbage faster." Fair. Agency without judgement is chaos. With it, leverage. The high-agency operator is not the one who ships the most. They are the one who ships the right thing fastest because they are closest to the person it is for. The Buggy Smart pivot to Verified Accreditation came from one parent saying "a list is not enough, I need to know it is actually true." That sentence rewrote the product. Customer pulls the trigger. Taste aims.
Two: "Klarna walked the AI back. They rehired humans." They did. And it proves the point. They overshot, customers pushed back, they adapted. The loop worked. The mistake was not the multiplier, it was thinking the multiplier removed the constraint. The constraint is taste and the customer. Always was.
Three: "In regulated orgs you cannot ship by Friday." True. So the move is not to pretend regulation does not exist. The move is small teams with customer access and ship-to-staging permission, no steerco. Stripe runs this way. So do most of the AI labs. So could most enterprises if they wanted to.
Four: "High-agency people burn out and leave." Also true. The org that loses them already lost. Build for them, or watch them ship for someone else.
The loop is older than 2024
I have been running this loop for fifteen-plus years. It looked different in different shops. At R/GA it was the prototype-led pitch. At MediaMonks it was a Web3 brand build that started as a sketch in a notebook. When I founded Burst it was a year of weekly experiments, most of which failed and a few of which mattered. Now solo, it is twenty-plus AI-native products live in the past year.
The reason high agency looks innate is that the people who have it have run the loop a thousand times. You do not hire for it. You set up an environment where it is allowed to compound. Every loop makes the next one faster, and the fear of deciding smaller. After enough reps, the decision is not a decision anymore. It is a reflex.
So what does high agency look like, for me, in 2026?
Being able to be the designer, the creative, the strategist, the builder, the writer, the product guy, the social guy all in one. The wearer of many hats, like a startup founder.
That is the bet I am making with my own time. Not "I can do everything." That would be silly. But the role I now occupy in any project is closer to a one-person studio than a one-job specialist. The tools made it possible. The years of running the loop made it work.
Agency was always the asset
The next decade does not belong to the people with the best tools. Everyone has the same tools. It belongs to the people who do not wait to be told. Who see the gap, decide without permission, and ship.
That was always true. AI just removed the alibi.
The full argument, eighteen slides: The High Agency Multiplier.
