The Manager Layer is Next
There's a pattern I keep seeing across organizations. A cross-team initiative needs a decision. You ask the manager. They say, "Let me go chat with the team about that." Two meetings later, you get a non-answer wrapped in process. Nothing ships.
That's not a bad manager. That's a job description that never required anything more than information routing, and AI is about to make that job structurally impossible to justify. The real question is which managers are actually load-bearing.
The scale-up era created management debt
Between 2019 and 2022, engineering organizations hired fast and structured even faster. Every team needed a manager. Every manager needed a skip-level one rung above. The result was coordination layers that produced more meetings than output. Elena Verna makes the case that AI is making it possible for senior ICs (individual contributors -- engineers who build, not manage) to drive company-level impact solo, which reframes everything.
Some of that structure was justified by the real complexity of coordinating large teams. But a lot of it was simpler than that. When teams are big enough, someone has to handle the status updates, the ticket grooming, the "what's the status on this?" Slack messages, the ferrying of information between levels. That routing is the work AI is absorbing first, well before it touches the strategic decisions. Remove it, and the pure people manager's job description gets very short very quickly.
I traced the IC side of this shift in October: the senior IC role concentrated, the junior pipeline contracted, and the manager span on paper didn't widen because the leverage per IC went up instead. This post is about what that did to the manager layer above the ICs.
The conduit manager can't survive
The kind of manager AI is now exposing has one defining trait: they couldn't take a position. Not because they were conflict-averse, though sometimes that too, but because they genuinely didn't have enough technical or business context to form one. Every question that crossed their desk got answered with a team consultation. They were a conduit.
This isn't a character flaw, it's a job description. The conduit manager isn't lazy and is often working very hard, but the work is information arbitrage. They had organizational access their reports didn't, and they used it to route. AI is dissolving that asymmetry: as agents get better at pulling context from Slack threads, docs, and tickets, and reports get business context directly, the argument for the coordination layer collapses.
Good people management, running performance conversations, developing careers, de-escalating team conflict, is real work that AI isn't absorbing. But it doesn't fill a full management headcount slot the way coordination did. When the coordination overhead goes away, what remains doesn't justify the layer.
What actually survives
The job forks hard, and the version that survives is the player-coach. Not the hollow "I stay technical" version that senior managers say and don't mean, but the literal one: someone who holds the technical and business context deeply enough to make hard calls when priorities conflict, represent their team credibly in any room, and decide without running it by six people first.
The best engineering leadership rooms are the ones where everyone in them is technically grounded. Alignment on org positioning happens fast because nobody needs to step out and check with their team to understand what's technically possible. Tradeoffs land on the spot because each person holds both sides of the equation. That kind of alignment is nearly impossible to replicate with a layer of coordinators in between, and it's what AI can't replace. The best directors are the ones who could succeed in either an IC or EM role. That's what makes them load-bearing. The coordination-only layer isn't.
The career question nobody's answering honestly
Most of the conversation about AI and management stays at "will managers be replaced?" Director and above jobs survive if the person can make real technical bets. The CIO Dive piece on engineering role shifts frames it as "AI saves time, then review explodes," and as I've written before, the review bottleneck is real. The deeper implication is that the review shift moves judgment upstream, to the person deciding what to build and why. That work is technical leadership.
The manager level gets squeezed hardest because it was always closest to the coordination-and-reporting work that AI absorbs. When the middle compresses, the layer above gets leaner, more technical, and less about headcount.
The IC path, meanwhile, is getting more interesting than it has been in years. Senior engineers who can direct agents, own entire product surfaces, and ship like a small team are valuable in a completely new way. As I wrote in AI Is Redefining the Senior Engineer, the ceiling for what one person can build is rising fast. The Engineers Codex is already tracking how misaligned incentives push engineers toward the wrong behaviors: gaming token counts, manufacturing complexity for promo cycles. Better metrics won't fix that. Fewer layers between intention and output will.
I keep my own IC muscles sharp because the player-coach is the only version of engineering leadership I think holds up through this. The people who thrive on the other side could always have been staff engineers if the career ladder had gone differently. They went into management for impact at scale, and never to stop doing the technical work.
If you're a manager today, the real question is whether you can credibly hold both sides. The scale-up era made it easy to build a career in the middle. The current era is a lot less forgiving about what you can actually do.