I lived as an AI rookie for two years.

There's a specific kind of quiet that shows up in a founder's week when the old playbook stops working. Not a crisis — just a slow, low hum that says the thing you were good at is no longer the thing the work needs. That hum started for me around late 2023, and it took me two years to name it honestly.

For 25 years I had a straight answer to almost every question a client could bring me. Sitemap? Here. Estimate? Here. What a launch week actually looks like at 2am? Here. Then, gradually, the questions changed. They stopped being about pages and estimates and started being about workflows, judgment, and where an AI teammate should — and should not — sit inside a small company. I did not have straight answers. I had to be a beginner again.

The first year was mostly humiliation.

I built things that did not ship. I over-trusted models on tasks they were bad at, and under-trusted them on tasks they were quietly excellent at. I spent one entire month designing a workflow that a good prompt and a spreadsheet could have replaced in an afternoon. Twice I told a client I would 'wire in AI' with a straight face, meaning I had no idea what I was going to wire in, or where.

"Being a rookie again, at 50, in front of my own team — that was the actual work. The tools were the easy part."

What actually changed the second year.

The shift wasn't a better model. It was a smaller question. Instead of asking, 'how do we adopt AI?', I started asking, 'which one workflow, if it ran itself for a week, would give me back a full day?' That question is boring on purpose. It forces you off the roadmap and into the calendar. Nine times out of ten, the answer is not glamorous — it's an intake form, a rewrite of a status update, a triage of an inbox.

• The workflow has to be one you already do every week, not one you wish you did.
• The output has to be something a human is going to read and sign off on — not a fully autonomous decision.
• The savings have to be measured in the calendar, not in tokens or in vibes.

What I'd tell the me of two years ago.

Don't hire the model as a coworker on day one. Hire it as an intern who is very fast and very literal, and give it the boring 20% of your week first. If it can't handle that, it definitely can't handle the interesting 20%. And write down what it got wrong. That log becomes your actual playbook — much more than any framework you'll read online.

Two years in, I don't feel like I've caught up. I feel like I finally know what game I'm playing. The next letter is about the three workflows I trust with my Monday morning, and the one I've quietly retired.