Managers Matter More Now, Not Less

There's a comfortable narrative going around right now: AI writes the code, so the engineers who manage other engineers are the obvious overhead to trim. More output per head, fewer heads needed to coordinate that output. It's a clean story, and it's wrong.

If anything, this is the most important moment to have good managers in the room.

What managers actually do

The job has never really been about reviewing pull requests or sitting in standups. A manager's actual function in an organization is to distill, disseminate, and expand information across the team and across the seams between teams. They translate strategy into work, advocate for their people, hold the domain context that lets a group make good calls under uncertainty, and notice the work that's quietly falling through the cracks because no one's job description quite covers it.

That last one matters more than most people realize. Large organizations don't fail at the things written down in OKRs. They fail at the things in between — the integration points, the handoffs, the half-owned systems. Managers who've been around long enough to see those seams are the ones who keep the org from quietly breaking itself.

None of that goes away when AI enters the picture. It gets harder.

Why AI raises the bar instead of lowering it

When your team is shipping more code, faster, with agents handling larger chunks of the work, the bottleneck shifts. It's no longer “can we write this?” It's “are we writing the right thing, in the right shape, with judgment we can trust?” That's a management problem, not a throughput problem.

Concretely, a manager whose team is using AI tools effectively now has to:

• Train their ICs on how to manage agent context — what to give the model, what to withhold, how to structure a project so an agent can actually be useful in it.
• Hire for judgment, not just velocity. The cost of a bad hire goes up when that person is now amplified by tools that will confidently produce a thousand lines of plausible-looking nonsense.
• Show the team — by doing, not by memo — what good use of these tools looks like. Where they help, where they hurt, where they need a human in the loop.
• Build the taste and the review culture that catches the failure modes specific to AI-assisted work.

A hands-off manager who hasn't touched the tools can't do any of this. A hands-on manager who has gotten genuinely familiar with where AI shines and where it falls over is a force multiplier in a way the org chart doesn't capture. They're the ones turning a fast team into a fast team that's also pointed in the right direction.

The strategic mistake

Some companies are pattern-matching on “AI = fewer engineers needed” and reaching for headcount cuts without asking the harder question: who on this team has actually adapted? Who's been using these tools effectively? Whose pre-AI skills — systems thinking, domain expertise, a nose for organizational seams — translate forward and compound with the new tooling?

The answer is rarely correlated with title or tenure in the way layoff spreadsheets assume. Some senior people have leaned in hard and become dramatically more effective. Some have not. Same for ICs, same for managers. Cutting by org chart instead of by adaptation is how you end up with a team that's smaller, cheaper, and quietly worse — because you kept the people whose work is most easily replicated by AI and lost the people who knew how to direct it.

The companies that come out of this period strongest will be the ones that took the time to figure out who in their organization has built real fluency with these tools, who can take their accumulated judgment and apply it to a new way of working, and who's worth keeping precisely because they make everyone around them better. That's a manager-shaped role, even when the person filling it isn't formally a manager.

The actual ask

If you're an IC: get fluent with the tools, but also pay attention to the people on your team who are good at the connective tissue work. That's a skill, and it's about to be more valuable, not less.

If you're a manager: don't wait for permission to get hands-on with this stuff. The managers who'll matter in two years are the ones who can speak credibly about agent workflows because they've actually run them, not because they read a thinkpiece about them.

And if you're making decisions about who stays and who goes — slow down and ask the right question. Not “do we still need this layer?” but “who in this layer has already adapted, and what would we lose if they walked out the door?”