Your best engineer is the one you are most likely to lose
High performers do not need less management. They need a different kind, and most of them are getting none at all. What I do differently with the people who carry the team.
The most productive engineer I ever managed almost quit because I was not paying attention to her. Not because I was ignoring her. Because I had decided, without ever saying it, that she did not need me.
She was shipping. Her reviews were clean. She never brought problems to our 1:1s, so our 1:1s got shorter, and then they got skipped when I was busy with the people who were struggling. When she finally told me she had been interviewing, the reason she gave was that she could not see what the next year looked like, and nobody had asked her what she wanted it to look like.
That was a management failure, and it was mine.
The time goes to the wrong people
New managers spend their attention where the pain is. The engineer who is behind, the conflict in the review queue, the project that is late. That is understandable and it is backwards. The people who are quietly carrying the team are the ones whose departure would hurt most, and they are the ones getting the least of you.
I now split my attention deliberately. The strongest people on the team get the same fixed 1:1 time as everyone else, and they get more of my thinking between meetings: which problems I route to them, which meetings I bring them into, whose name I mention when leadership asks who should own something.
What they actually want
I have asked a lot of high performers what they want from a manager. The answers cluster.
They want problems that are hard enough. The fastest way to lose one is to keep handing them the same kind of work because they are good at it. Being good at it is why it has stopped being interesting.
They want their scope to grow ahead of their title. Promotion cycles are slow and the work is not. If someone is operating at the next level, let them own something at the next level now, visibly, and make the promotion a formality.
They want the friction removed. A high performer blocked on a decision for a week is the most expensive idle time on the team. When one of them tells me they are stuck on something outside their control, that becomes my most urgent item.
They want honest feedback, and they get less of it than anyone. Nobody wants to criticize the person carrying the team. So they hear "great work" for years and nothing about the way they run over people in design reviews. Then that habit blocks their promotion and they find out from a calibration outcome instead of from you.
The thing nobody says
Some high performers are hard to work with, and the team knows it before you do. I have had to tell an engineer whose output was double anyone else's that the way he handled disagreement was costing the team more than his output was worth. It was the most uncomfortable feedback conversation of my career and it was overdue by about a year.
He took it better than I expected, because it was specific and because I had the examples. What he said afterward stuck with me: nobody had told him, so he assumed it was fine.
What AI changed
When the tools arrived, the definition of a high performer moved under everyone. A few engineers who had been the fastest coders on the team found that speed was no longer the scarce thing. The scarce thing was judgment: knowing what to build, catching what the tool got wrong, and explaining a change to someone who did not write it.
The high performers who adapted fastest were the ones who had always been strong on judgment and were now free of the typing. The ones who struggled were the ones whose identity was built on producing code by hand. I spent more 1:1 time with that second group than with anyone else that year, because their standing on the team was changing and they could feel it.
What I do now
Every quarter I write down, for each person, what would happen if they resigned tomorrow. For the people whose name makes me wince, I check three things: is the work still hard, is their scope ahead of their title, and when was the last time I told them something they did not want to hear. If any of those is wrong, that is the next 1:1.