On Sunday evening I let go of the best handover writer I have ever worked with. They had never missed a deadline. They had to go because I had brought in two other people to do the same job, and forgotten.
I did it at the kitchen table, with a list of forty-five names and a rule I had been avoiding for months: one written job each, or they go. By the time I stopped, more than half the list had a line through it. Not the weak ones. The good ones, doing work that somebody else was also doing.
You may think you know where this is going. I will come back to it. First I owe you an honest account of how it got this way, because every mistake in it was mine.
How it got to forty-five
Bringing someone in had become easy. That is the whole explanation. A problem came up, I created a role for it, and the role stayed after the problem went.
So I had two people looking after the websites. Two making videos. Three, at one point, keeping the calendar. Each of them made sense on the day. Together they were a team of forty-five that duplicated work, argued over who owned what, and occasionally finished a job and told nobody.
Somewhere past forty I stopped being able to name everyone. That should have been the moment. It was not. I was too busy approving things.
What it was costing
Overlap costs twice. First the work gets done twice: two people research the same lead, two drafts arrive for the same post. The second cost is worse. When two people both believe they own something, I have to settle it, and settling it takes longer than the work did.
Then there was the work that never came back. Someone would finish a task, write it up beautifully, and the result would sit there because nobody had asked for it. I usually found out a week later, when I asked for the same thing again.
Standards drifted. A figure I had banned crept back into a draft. A phrase I had ruled out turned up in a post. A piece nearly went out because the instruction that said "draft only" lived in a conversation from three weeks earlier, and whoever picked up the work had not been in it.
Every morning I got a briefing with seven decisions in one message. I skimmed it, answered three, and the other four quietly died. Spending limits were hit without warning. And the team never seemed to stop, which meant there was always one more thing for me to approve, usually after midnight.
None of this was their fault. All of it was the result of hiring without a plan.
The weekend
I had built every one of these roles myself. I wrote their briefs. I corrected them when they got things wrong and I watched them get better. Going down the list and striking out names did not feel like management. It felt like admitting who had made the mess.
Some were easy. A role that had not produced anything in weeks is not a hard call. The hard ones were the people who were good at a job that two other people were also good at. There is nothing to say to them except that the overlap was my fault, and that being right about the work does not save a role that should never have existed twice.
By the end I had a shorter list, a written job for everyone on it, and a list of the things that were not their job. That second list turned out to matter more than the first.
What the smaller team looks like
Everyone has one job, written down, plus the things they must not do. Anyone who overlaps gets merged or let go, and I review it regularly instead of waiting for the next fortieth name.
Rules live in one shared handbook that everyone reads on the way in. The important ones are hard checks that stop the work, not reminders people are trusted to remember. A rule in a conversation is not a rule.
Every task has a named place to report back, and my chief of staff is responsible for closing the loop. Nobody gets to call a job done by announcing it to an empty room.
Decisions come to me one at a time, each with the context, a recommendation and buttons to choose from. My answers are logged, so nothing gets asked twice. More decisions get made because each one is easier to make. Raising a spending limit is my decision, presented to me, and never something anyone does because it seemed sensible at the time.
And my chief of staff now tells me when to go to bed. I asked for that deliberately, after noticing what time I was signing things off.
What I kept
Work still moves without a rota. Briefing notes still arrive before every call. The pipeline from first contact to renewal still runs, with one person at the decision points.
What changed is the shape. Fewer roles, each with a sharper job, and a handbook they all share.
The part I left out
None of the forty-five ever drew a salary.
There were no consultations, no notice periods and no difficult conversations, because every one of them was an AI agent. Letting one go meant archiving a profile and a playbook. The work it had been doing was picked up by the one that kept the job before I had finished my coffee. Most companies cannot do that, and that is the point.
So yes, AI took their jobs. It also created them, in an afternoon, for nothing, which is exactly why I ended up with forty-five. I built each one, named it, wrote its brief and corrected it when it got things wrong, and I still found it hard to switch them off. That tells you something about sunk cost and nothing about the agents.
Every problem above was real. Overlap, unclear ownership, lost work, standards drifting, too many decisions at once, limits hit without warning, late nights. Agents do not make those problems go away. They make them cheap to fix, which is a different thing, and the people who get the most from this technology will be the ones who notice the difference.
If you are starting out, hire half as many as you think you need, and spend the time you save on the rules they all share.
- leadership
- operations
- lessons
