I’ve worked in AI for more than ten years, and I still can’t stop talking about it.

Context Is King

The Smartest AI in the World Still Doesn’t Know What Your Company Does

Most people have worked out that AI does better work when you tell it about yourself. Plenty of us keep a little file of notes to paste in. I had one too, two years ago. Doing that for a whole company is harder.

In 2024 I started building a brain for the company I worked at, so its AI tools would know how the business ran. Nobody asked me to. More than a year later its engineers came to the same design by themselves and asked me to hand it over.

A brain like that wears down all the time, and it’s easy to miss. Often it’s just that something the AI was told stops being true. Researchers have found stranger ways it happens, like the effect they call lost in the middle, where the AI does worse with the middle of a long document than with either end. I’ve been tracking nine of these.

I haven’t seen an AI platform judge which of a company’s facts to throw away. Nobody knows where AI goes over the long run, me included. For now, somebody inside the company has to decide what its AI gets told and keep checking it’s still right. I’ve got plenty of that wrong, and I’ll tell you where.

You’ll leave with

• Nine ways a company’s AI brain wears down, and what I do about each one.
• Why an AI that remembers you still doesn’t know your company.
• Where I’d start tomorrow if I had to build a company brain again with no team.

Ask AI What to Stop

The AI System I Built Takes Work Off My List Every Morning

Most of what we ask AI to do adds to the pile. It writes another draft, or another list of things we could do next, and it’s very good at that. What I needed help with was deciding what not to do.

So in 2026 I built myself a personal operating system to do that. Every task gets checked against the targets the business actually has. It catches the work I pick up because I like helping, when it isn’t mine to do. Each day I get three things to do, and a list of what I’ve let go, so I don’t feel guilty about it.

It doesn’t try to remember everything. It resets every morning and starts again from what matters to the business now. One part of it is only there to protect it from me. When I’m stressed and try to change what it checks, it won’t let me.

You’ll leave with

• The questions my system asks about every task before it gets anywhere near my day.
• What goes on the list of things I let go, and why I keep it.
• How to get the AI you already use to tell you what to stop.

Why Hasn’t AI Changed the Work?

It Changes When You Build an Agent With the People Who Do the Job

In a lot of companies, a few people have made AI part of how they work. Most tried it, got something generic or something that wasn’t true, and went back to how they did it before. A habit doesn’t move because the company asks. A rollout gets you logins and finished courses.

On my first day as a marketing director, I was asked who my customer was. I said the people buying from us and was told I was wrong. It was the Chief Commercial Officer, in the same building. That stuck with me because it was right, and I think AI inside a company has its first customer in the building too, the colleague who’s going to use it.

When I built AI tools inside a company, I built them for the people who’d use them. Each one got an owner, and colleagues with the same job switched it on without being taught. Now I go a step further. I pick one job somebody does every week and build an agent with them, so the person who shows it to colleagues is someone whose job they recognise.

You’ll leave with

• The two numbers I’d measure instead of logins, and why fewer rounds before sign-off means better work.
• How to pick the first job to build an agent for.
• Who inside the building to build it with, and who should own it once it works.