I listened to David Senra ’s conversation with Nik Storonsky, the founder of Revolut, released on October 25. He’s one of the toughest entrepreneurs I’ve heard, and I’d argue with some of his views. But three of his ideas are worth taking home, especially for anyone trying to build an AI-first company.
A rule picks the products
Revolut runs product development as a portfolio of bets. A team of about ten people, with back-end and front-end developers, a designer, a product owner and sometimes a data scientist, gets 9 to 18 months to build a new product. After launch, one number decides its future: growth in gross profit.
Over ten years Revolut has launched about 50 of these bets, so it knows what a good curve looks like. Its best bets grow 10-20% month on month. A new bet that lands in the bottom 25% of all past bets is shut down. Bets near the top get a second team, then a third, and eventually become a business line of their own.
What struck me most is why he built it that way. He says he doesn’t trust his own intuition, because it was wrong too many times. Founders rarely say that out loud.
I have written before about the small group of people inside a company who create most of the new value: they propose a hypothesis, test it with real money and real customers, and then scale it or drop it. In my experience, launching ideas is the easy part. Closing the ones that don’t grow, honestly and on time, is where most companies struggle. A rule agreed in advance takes ego out of that decision.
Small teams and the AI employee
Ten people per bet was the right size for the last decade. Storonsky now thinks it should be two or three at most, because one developer with AI can cover the front end, the back end and even data science.
The numbers he gives for the rest of the company are just as direct. Customer support used to take more than 5,000 people, and about 90% of it is now automated. Engineers work about 30% faster. The goal he describes is an AI employee with its own email and Slack account, access to back-office systems, and a seat on calls.
My favorite detail is about design. One team stopped drawing screens in Figma. They gave a model their design system at the code level and now ask it to build live interfaces you can click through on a laptop or a phone simulator. He says it is three to four times faster than designing frames, and he is now pushing the method to every designer.
We went through a similar change ourselves, moving from detailed written specifications to prototypes people can touch. It changes the discussion: everyone reacts to the same working screen instead of reading a document in their own way.
The founder’s job is to spread what works
Storonsky describes the company as a machine. You build it, hit a bottleneck, fix it, and hit the next one: first speed, then regulation, then hiring. When one team finds a faster way to work, his job is to make it the standard for everyone, quickly. The designers are his example.
Senra compared this to Andrew Carnegie, who looked for better methods wherever they were, including in British factories, and adopted them before his competitors did.
The principle carries over well to AI. Inside large companies I keep seeing the same pattern: one team quietly finds a way to work twice as fast with AI, and the rest of the organization never hears about it. Spreading that method is the harder and more valuable part of the job.
Where I would run it differently
He lets go of weak performers and average ones too. He says empathy comes second to winning, and that working weekends is a must if you want to survive. He points to the co-working space where he started: of 140 companies there, two survived ten years, and they all worked weekends.
For Revolut, racing to build a global bank, that approach may work. I paid a high price to learn a different lesson: a business needs people who work best together, and a collection of the strongest individuals is not enough. A machine can break from overheating as well as from weak parts.
Where I agree completely
Experienced executives rarely work out. He says that in Revolut’s first five years he hired more than 50 of them and kept two. The best people have already built what they wanted and aren’t looking for a job. So he hires people aged 25 to 30 whose career progress puts them in the top 1% of their peers, and grows them inside.
My best results also came from our own people who were given a task far above their level at the time.
What an AI-first company looks like outside a bank
Put together, it is a machine of small teams, decisions made by numbers, and AI in place of layers of management. The founder spends his time on bottlenecks and on making sure that what worked for one team quickly works for all.
Earlier this year I wrote about the 20x company, where the biggest barrier to growth with AI is the person who stays in every approval loop. Revolut shows what that looks like when someone builds it, in one of the most regulated industries there is.
For most of the companies I work with, the first steps are smaller and very practical:
Treat AI initiatives as a portfolio, with a closing rule agreed before launch.
Staff the next new team with two or three people and AI, and see what breaks.
Pick one method a team has already proven and make it everyone’s default within a month.
How to treat people is a choice each company makes for itself. The machine side is available to everyone.
The full conversation: