Everyone has decided managers are the most endangered species, that the coordinator class gets automated away first. I disagree.

Pure process-managers who exist only to relay information and track Jira tickets have a problem, but that’s not new. And in the agent era you have to produce, ship, and cover more surface area than before. The pure toll-booth manager is, yes, cooked — and the org-chart flattening happening across big tech right now is real. It’s just hitting that manager, not the one this essay is about.

The manager who actually knows how to manage is first in line to extract leverage from AI agents. This is not because they’re senior or because they have budget. It’s because they’ve already built the muscle that determines who wins here, and most individual contributors haven’t had to build it yet.

The Muscle Most ICs Don’t Have

Using AI agents at full leverage requires you to release something most people hold more tightly than they realize: control over how the work gets done.

Not the outcome. The how.

A good manager has been practicing this for years. Their team wrote the customer email, not in the words the manager would have chosen. The engineer built the feature, not quite the architecture the manager pictured. The analyst built the deck in a slightly different structure. And the manager let it ship. Not because they didn’t care. Because they understood the trade: leverage means the work doesn’t have to look like you for it to be good.

That muscle takes time to develop. Most ICs have never had to touch it.

This Isn’t About Settling

I don’t want the argument to be misread, so I’ll be direct.

I watched a senior engineer spend three hours wrestling an AI coding agent into submission, rewriting its outputs line by line until the variable names matched his convention, the comments sounded like him, the structure felt familiar. The agent was producing correct, clean code. He was making it his. By the end, he’d have been faster doing it himself, and he knew it. He kept going anyway. That’s not a skill problem. That’s an identity problem: the output is too bound to how he does it, not just what gets done.

The argument is not “accept low-quality work.” The argument is: be willing to accept output that doesn’t match your exact voice, which frequently means it’s better than what you’d have produced yourself. The craft is knowing when variation matters and when it doesn’t. If the email accomplishes the goal, the different phrasing is fine. If the code ships clean, the different naming convention is fine. You intervene when it actually matters (and you have to know when that is), and you let it go when it doesn’t.

Most managers have been making that call every single day.

The First Year of Managing, Again

That same identity problem shows up in every new manager. The first year of managing is hard for exactly this reason. You have to stop measuring success by whether the output looks like you built it. You have to learn to care about the result without needing credit for the method.

The managers who’ve gotten good at their jobs have already wrestled through that transition. They’ve accepted that the team’s output is the output, not a pale copy of what they would have made. They’ve learned to define the outcome and stay out of the way on the rest.

I have a name for the disposition this requires: outcome ownership without method attachment. You’re fully accountable for the result. You’re not precious about the path.

That is the exact disposition AI agents require.

The Playbook Transfers Directly

The playbook for working with agents is almost identical to the playbook for managing people: define what done looks like, set the constraints that actually matter, stay out of the way on everything else, and course-correct when something actually goes wrong. Managers have been running that loop for years. They have the reps.

The IC who can’t hand off the keyboard is going to grind manually while peers 10x their output. The manager who’s spent years practicing outcome ownership without method attachment is going to look like they have an unfair advantage because they built the only skill that transfers cleanly to the agent era.

That’s not a management skill anymore. It’s the most important skill in the building. The managers figured it out first (whether they meant to or not).