The handoff was always the job

Everyone is suddenly learning to manage AI agents. Write a clear spec. Break the work into small pieces. Review every output. Never assume it understood you. Watch out for the confident wrong answer.

I've been doing that for years. We just called it directing contract developers.

I spent about five years as a product owner over a portfolio of enterprise apps used by more than 5,000 people. The developers weren't my direct reports. I couldn't tap anyone on the shoulder or walk over and explain it again. Everything I wanted had to survive the trip through a written spec. What came back was exactly what I asked for, which is not the same thing as what I meant.

You learn a few things in that seat.

A vague requirement doesn't come back as a question. It comes back as confident, well built, completely wrong work.

The review is the job. It isn't the thing you do after the job.

Trust is earned per task, not granted once.

Every one of those habits transferred to working with AI. I didn't have to change a single one of them.

We keep talking about this like it's a tooling problem, like the people who learn the syntax fastest will come out ahead. I don't think that's it. The tools are getting easier every month. What isn't getting easier is knowing how to describe what you want to something that can't read your mind, and then having the discipline to actually check.

The people who end up good at this won't be the ones who learned the tools first. It'll be the ones who already knew how to hand off work.