The job of a manager used to be, in large part, coordination. Knowing what everyone is working on, relaying it upward, making sure the pieces fit. AI is collapsing the value of coordination. If coordination was your primary value as a manager, you have a problem.
I had two experiences that made me a much better manager.
The first turns out to be rarer and more valuable than I realized at the time. I had two great managers early in my career. One was a classic in the trenches lead — deeply in the detail, always available, always knew the answer. The other was almost the opposite. She set direction, cleared the path, and then got out of the way. From the first I learned how to be useful in the moment. From the second I learned how to build something that works without you in the room.
The second was that during my time at Culture Amp, I ended up leading the Sales and Marketing organization. I'm an engineer, so that's not exactly a natural transition. But it turned out to be the most formative management experience I've had. When you can't do the job yourself, you can't fall back on being the expert in the room. You have to focus on the system: are the right people in the right roles, do they have what they need, is the feedback loop working? You stop managing the work and start managing the environment the work happens in[1].
Both experiences pointed in the same direction: the best management I've seen, and the best management I've done, was never about coordinating work. It was about creating the right environment.
A lot more gardening than architecture.
This isn't new. But AI is changing and broadening the implications. AI is forcing every manager into that reality. ICs are now coordinating their own agents, shipping work in parallel, managing their own loops. When output scales to 100 PRs a day, you can't review all the work. You shouldn't be in every loop.
So the question becomes: what does the manager actually do?
You design the system and nudge it towards the right shape. What are the rubrics? What are the defaults? What gets flagged and what doesn't? When your people systems are a codebase, the manager's job is reviewing commits to the system, not the product. Managing that Marketing team as an engineer taught me what it feels like when you can't be the expert[2]. AI is doing the same thing to every manager now (and perhaps every IC, to a different extent).
None of this makes management easier. If anything, it raises the bar. That second manager I had, the one who set direction and got out of the way — she had to be far more deliberate about the environment she created precisely because she wasn't in the detail. That was the harder job all along. It's the only job left now.