AI made the work fast and cheap. Organising people did not get any easier.

In September I stood on the sideline while my ten-year-old and four of his mates played a five-a-side soccer tournament. It ran like the Champions League, with six teams and no coaches. Nothing stopped a parent from being one.

Within minutes of the first game I could see what each of them was good at. Two could dribble. Two could strike the ball hard. One could defend. What I could not see was a team. All five chased the ball around one square metre, so nothing came down the left or the right. They were each playing for themselves.

In the two minutes between games I hesitated. Four of them were not my kid, so I asked first. "You guys are all awesome and that was a great game, but are you willing to listen to some feedback? It'll help you perform even better." My son was over the moon. Yes baba please, we want to win.

I did not give them a hundred things. This was not a one-man sport, so back each other up and pass. Stop bunching up. Spread out and come at the goal from the left, the right, and the centre. Aim for two things: have fun and try to win.

I also gave them positions. The hardest kicker went up front, the best stopper went in goal, and they swapped around from there. The bigger boy, a bit of a leader, became the unofficial captain. I told him to put the others where he thought they belonged.

I watched them win for the rest of the day. Nothing changed except where they stood and who got to decide that.

The dot is the same dot in both panels, a destination on the left and already moving on the right. Nothing about the five changed. Generated by Author using Claude.

. . .

I added no new skill that weekend. I only changed how they were organised: a position each, a clear idea of winning, one person allowed to place the others, and the freedom to pass. It got me thinking about my teams at work.

What AI changed at work

For years I worked on teams where the limit was how much work got done. Few people could build software well. So skill was what you hired and paid for.

Now I see one engineer with an AI tool do in an afternoon what used to take a small group a week. The skill did not go anywhere. It just stopped being rare.

I have not seen the work around the code get any cheaper. Deciding what to build is the same job it always was. So is checking that somebody else's code is right, except far more of it lands every day.

I no longer see the hold-up inside one person. I see it where one person's work passes to the next.

The pile-up got faster

Picture five engineers with AI tools piling onto one problem. Nobody decided they should all be on it.

Before AI I could spot this because it was slow. Two people building the same thing would run into each other within a day. One had to wait for the other.

I do not see anyone get stuck now. Five people can each build a full version of the same thing in one afternoon without getting in each other's way. Every version looks right.

I have been around teams that ended up with three versions of the same small piece of code, two clashing changes to one database table, and a full set of automated tests for a part of the system that was about to be deleted. All of it ran. Every change had a tidy write-up and a green tick from the checks. Nobody had decided who should do which part.

The work I worry about is the kind that is wrong but looks right. Wrong code that looks wrong gets caught in ten seconds. Wrong code that looks right only gets caught when someone checks it with enough time and background. The people checking now have far more to get through and no extra hours. Stack Overflow is a website where programmers ask and answer each other's questions. In its 2025 Developer Survey, 66% of developers said their top frustration was AI answers that are almost right, but not quite.

I have never seen the numbers look this good. More code is saved, more changes go up for review, and more tasks get ticked off. But the usual speed measures count the pile-up, not the result. Google runs a yearly study of software teams called DORA (DevOps Research and Assessment). Its 2024 report found that where AI use climbed without the right habits around it, delivery speed fell 1.5%. Stability, meaning how often changes go out without breaking something, fell 7.2%. DORA's later reports say much the same. AI makes a team more of whatever it already is.

What running a team means now

I think running a team is now mostly decisions that have nothing to do with how well anyone codes.

The first is who owns which part, written down, not assumed. "We all own everything" was fine when work came slowly.

Next is what winning looks like this quarter, in one sentence the team can say back without opening a document. Twelve goals is the same as none.

Strengths need naming honestly too. Somebody on your team is the best at tracking down bugs. People are too polite to say it, so work goes to whoever is free, not whoever is best.

Leadership has to be clear too. Not a title, but the person the team already listens to, with the room to lead.

The last is how the work moves. Big pieces handed over at the end are where AI-made work piles up, waiting for someone to check it. Small pieces, worked on together and checked and joined up as the team goes, keep it moving. AI made doing the work cheap. It did not make checking it and fitting it together cheap.

Where the soccer story falls short

I had the easiest coaching job there is: one ball, twenty minutes, a whistle, and a scoreboard everybody could see.

I have never worked anywhere with a whistle. The quarter ends and the work keeps going. You move people around mid-game, which is a different job from doing it between games.

I also had the easiest people to move. Those five had nothing riding on it but the next match. A senior engineer who has looked after one system for three years does not hand half of it over because somebody drew a diagram.

And all the evidence I have is one weekend: five boys, one tournament, one result. What I trust is not the winning. It is that I could see exactly what was missing and none of it was ability.

AI will keep making the work faster and cheaper. On its own, that does not make a team better. Those boys could already play. What changed was that each one had a position, they all knew what they were aiming for, one of them placed the others, and they passed the ball.

A team at work needs the same. Clear roles and a clear goal come first. Then a self-sufficient team with the right leadership around it. The product owner sets the problem worth solving, not the answer. The team swarms on it, which means working on it together on purpose, explores it and comes back with options. A scrum master or engineering manager coaches the team rather than directing it. The work then gets done together, in small pieces that are checked and joined up as the team goes, so nothing sits waiting at the end.

It is not always that simple at work. But that is where I would start.

Salam.

P.S. With chai and love.