AI won't replace founders

I spend my days building infrastructure for AI developers at Substrat, and trying to use AI to help alienated fathers at Anchor. Because of this, I get asked the same question roughly twice a week by other founders, usually over a bad game of tennis here in Marbella or a glass of wine. They want to know if software is going to start founding companies. The logic goes that if a model can write code, run marketing, and handle customer support, the human founder is suddenly redundant. I tell them no, but the truth is actually much more uncomfortable. AI will not replace the person who starts the company. What it replaces is almost everything else we used to associate with running one.

The middle layer

When Fredrik Bage and I started Mediapilot back in Stockholm, we grew it over roughly seven years and ended up creating more than a hundred jobs. I am proud of that number. But if I am honest with you, a massive percentage of those roles were essentially human APIs. We hired people to take information from a client, reformat it, hand it to a designer, take the result, check it against a brief, and pass it back. It was the middle layer of coordination. We called it management, or operations, or delivery. That entire layer is vanishing. A language model is perfectly happy acting as the router of information. It does not get tired, it does not need a one-to-one meeting to discuss its career progression, and it routes data flawlessly. Saying this out loud feels awful, especially in Europe where we measure corporate success by headcount. But denying it will not save those jobs.

You cannot automate the act of looking someone in the eye and promising them that a chaotic, half-broken idea is going to work.

When the middle layer disappears, three things are left for the founder: taste, accountability, and persuasion under uncertainty. Before Mediapilot, I started out doing sales at a PR agency on pure commission, against the advice of everyone around me. But Fredrik was there, and he told me I would be good at sales, so I jumped. That entire phase of my life was an exercise in persuasion. I had to convince people to buy something based on trust. AI cannot do this. A machine cannot convince an early employee to quit their safe job and join your unproven startup. It cannot persuade an angel investor to wire you their money. Persuasion requires shared human risk.

The burden of taste

Then there is taste. At Happeo, where I worked on the commercial side of intranet and employee experience software, the product succeeded because human beings made distinct, opinionated choices about how other human beings should interact at work. An AI can give you fifty variations of a dashboard, but it has no earthly idea which one actually feels right to a tired employee logging on at nine in the morning. Building Anchor is exactly the same. We are using AI to help fathers who are separated from their children, a situation fraught with intense pain and complex legal realities. The models handle the heavy data lifting, but the tone, the boundaries, and the specific way we support these men comes entirely from human taste. We have to decide what is helpful and what is just noise. The software cannot make that moral and aesthetic judgement.

No place to hide

The hardest part of this shift is accountability. If an AI writes a piece of code that crashes a client's system, the client does not sue the language model. They sue you. You are the one who goes bankrupt. You are the one who has to call the customer and apologise. The founder is the ultimate sink for risk. We are moving towards a world where companies consist of a founder with extreme taste, a highly automated infrastructure, and maybe three exceptional operators. The days of hiding behind a vast organisational chart are over. You are stripped down to your raw ability to make decisions and carry the consequences. Every year since around 2010, I have written down a life plan, rating my work, finances, family and friends, health, and personal time out of ten. I do it to keep myself accountable to reality. Running an AI-heavy company feels exactly like that annual review, but every single day.

I haven't got this entirely figured out. I know the transition is going to be messy, and I know that taking away the coordination work from humans removes the training ground where many of us learned how to run businesses in the first place. If nobody can get a junior role passing messages between departments, where do the founders of 2040 come from? I do not have a good answer for that. What I do know is that starting a company will remain a deeply human act of rebellion against the current state of the world. AI is a spectacular tool for execution. But it takes a person to look at the world, decide it needs changing, and put their own neck on the line to make it happen.