People Management in the Age of AI

The dominant story in tech right now is that flatter orgs, smaller teams, fewer people in the middle, and more “leverage” per leader are the path forward. AI is the engine being used to justify it: agents will do the work, the thinking goes, and managers are mostly overhead. I think the opposite is true. AI raises the value of management, not lowers it, and the orgs that figure this out first will have something rare in a couple of years: people who actually know how to lead other people.

So you can manage agents effectively, how difficult can managing people be?

Fair question, on its face. Agents really are easier in a specific sense. Yes, there’s context engineering. Yes, there’s evaluation, prompting, scaffolding. All real technical work. I’m not minimizing it. But agents don’t judge you. They don’t get quiet in standup because something happened at home. They don’t need to feel seen. They don’t watch their peers get promoted and wonder why they didn’t. They don’t burn out and quit at the worst possible moment for your roadmap.

Humans do all of that. That’s not accidental complexity you can engineer away. It’s the inherent complexity of working with people who have inner lives. Conflating the two is a category mistake. People who think managing agents has prepared them to manage humans are going to find out, expensively, that it hasn’t.

And once you’ve quietly accepted that category mistake, the rest of the received wisdom follows. If people problems are just an engineering problem in disguise, then of course you can delete the layer whose whole job was people problems.

”Kill middle management” kills the training program

I’m not here to defend middle management as an institution. Plenty of it is bloated. Plenty of middle managers are bad at the job. The honest pro-flat argument is real: the old system conscripted strong ICs into management because it was the only path to growth and comp, and produced a lot of reluctant, mediocre managers as a result. Build a strong IC ladder, the argument goes, and management becomes a choice for people who actually want it.

I get it. But here’s the part the argument waves away: management is a skill that has to be learned, and middle management was where you learned it. A first-time manager of three or four people inside a larger org has guardrails. Senior managers above them, peers around them, low-stakes mistakes to make. Take that away and the only management roles left are high-stakes ones nobody is qualified to step into cold. I don’t hand my kid a 10,000-piece Death Star Lego set and walk away. We build toward it together. It’s not that no kid could finish it solo, but the odds are bad enough that nobody sensible would set it up that way. That’s what middle management is: the staged build. You don’t get to skip it just because the trainees are eager.

And even ICs who never want to manage are still gated by managers: for promotions, for scope, for advocacy in calibration. If the management layer is thin and untrained, the IC ladder doesn’t save you. It rots from the support beam outward.

”The reward for good work is more work” is the burnout pipeline

That rot has a name on the IC side too. The pro-flat counter to titles and promotions is that they’re treadmill theater, that what people actually want is interesting work and good comp, not inflated titles. In a healthy market, maybe. In this market (layoffs, hiring freezes, “do more with less,” AI-productivity mandates from leadership decks), “no title inflation” has cashed out as same comp, more work, fewer people, indefinitely. That’s not a meritocratic utopia. That’s how you manufacture burnout at scale, and the reports are already rolling in.

People need legible progress markers. Titles aren’t the only kind, but “more responsibility, same everything else, forever” is not a marker. It’s an extraction strategy. And small teams have almost no slack for natural attrition. One person leaves, one goes on parental leave, one checks out, and the bus factor isn’t theoretical anymore. It’s Tuesday.

What stronger management actually looks like

If the diagnosis is that AI is being used to justify hollowing out the layer we most need, the prescription is pretty unglamorous. Three moves.

Keep the apprenticeship layer. A first-time manager with three or four reports inside a larger org is how managers get made. That role is cheap insurance against the high-stakes manager jobs above it being filled by people who learned on the way down. If you’ve deleted that layer to look efficient on an org chart, you’ve deferred a much bigger cost.

Narrow spans where the work is people-heavy. Wide spans are fine for coordinating senior ICs who don’t need much. They are corrosive when reports are early-career, when teams are forming, when there’s conflict, when there’s growth to invest in. The “one manager, twenty reports” picture is a budget decision dressed up as a structural one. Call it that out loud.

Treat management as a craft, not a default promotion. A real IC ladder is genuinely good. So is a real management ladder, with its own development track, its own peer bench, its own legible way of getting better at the job. You should be able to choose management deliberately and grow inside it without ever managing more headcount than you can actually carry.

None of this is at war with AI. Use the agents. Build the tooling. Cut the meetings that exist only because nobody trusts the docs. But don’t confuse a productivity unlock at the IC layer for a reason to delete the layer above it. The world is still messy because humans are still messy, and you can’t prompt-engineer your way out of that. The orgs that internalize this first won’t be the ones with the flattest org chart in 2028. They’ll be the ones still shipping, because they kept the people who know how to lead other people.

I would love to know how you are thinking about management in the age of AI. Share your insights!