The average engineering manager spends 17.9 hours a week in meetings. In 2024, managers made up 29% of all layoffs. Job postings for management roles are down 40% since 2022. Meta’s new applied AI engineering organization is running at a 50-to-1 engineer-to-manager ratio. And Gartner projects that one in five businesses will use AI to flatten their org structure, eliminating over half of current middle management positions.
These aren’t unrelated data points. They’re the outline of the same story: the engineering manager role is compressing — and the compression is accelerating faster than most people in the role are willing to admit.
This has happened before. Not to managers — to engineers. When cloud infrastructure matured, the walls between frontend, backend, and ops dissolved. The “full-stack engineer” wasn’t a new job category. It was the market repricing coordination. The cost of keeping specialists siloed became higher than the cost of one person knowing more. Role after role — DevOps, data engineering, QA — got absorbed into adjacent responsibility, not because companies wanted to squeeze headcount, but because the tools made it possible and the abstraction layers made specialization expensive.
The same logic is now being applied one level up the org chart. The question isn’t whether it will happen. The question is whether you’re positioned for it when it does.
TL;DR
- The EM role is compressing the same way the IC role did when “full-stack” became the expectation
- Agents are absorbing the coordination overhead that has always justified management layers
- Software design instincts are back on the critical path — not as a nice-to-have, but as a core differentiator
- 50 direct reports isn’t a fever dream — Meta’s applied AI engineering division is already running at that ratio
- The EMs who thrive will be the ones who lean into technical depth and learn to manage humans and agents simultaneously
What Agents Are Actually Replacing
Be honest about how you spend your week. The average engineering manager spends 17.9 hours per week in meetings — 7 more than ICs — and is left with just 10.4 hours of focus time. That’s not a calendar problem. That’s a structural one. The majority of those 17.9 hours are coordination: syncing on status, triaging priorities, running rituals, unblocking cross-team dependencies, reviewing output for process compliance. It even has a name — the “coordination tax” — and it compounds with company size. Engineers at medium and large organizations spend 3.2 more hours per week in meetings than their counterparts at small companies.
Agents do this work well. They don’t lose track of context between Mondays. They don’t need a standup to know where things stand. They can monitor progress, surface blockers, draft the status update, and flag the dependency risk before you’ve had your first coffee. Over 57% of organizations already have agents running in production, and the fastest-growing use cases are in exactly the kind of repeatable workflow and coordination work that fills an EM’s week.
This doesn’t mean the EM role disappears. It means the EM role compresses — and what’s left after compression is the part that always required judgment.
The coordination tax is getting automated. What remains is the work that was always the real job: technical judgment, people development, and the ability to see around corners.
Why Software Design Skills Are Back on the Critical Path
Here’s a pattern that plays out in a lot of engineering orgs: a strong IC becomes an EM, and within eighteen months, they’ve stopped reviewing designs with any real depth. They’re too busy. The calendar filled up. They trust their team. And gradually, their technical instincts go from sharp to soft.
In a world where coordination overhead was the dominant cost of management, this was an acceptable trade-off. The EM’s leverage came from process fluency and organizational navigation. Technical depth was nice to have.
That calculus is flipping.
90% of engineering teams are now using AI coding tools, up from 61% just a year ago, and 62% of respondents report at least a 25% productivity increase. As agents absorb the coordination layer, the EM’s remaining value concentrates in the places agents can’t go: architectural judgment, design review that catches the thing that technically works but will be a nightmare to maintain, recognizing when a team is building the right thing the wrong way. These are pattern-matching skills built on deep technical experience. You can’t delegate them to a model. You can’t recover them quickly if you’ve let them atrophy.
The EMs who have kept one foot in the technical work — who still review system designs with genuine opinions, who can sketch an alternative approach on a whiteboard — are going to have an enormous advantage over the next few years.
If you haven’t written a line of code or reviewed a non-trivial design in the last six months, that’s not a badge of seniority. That’s a gap that’s getting more expensive to have.
The 50-Report Manager Isn’t Crazy
The conventional wisdom on span of control has been remarkably stable for decades. Six to ten reports is the sweet spot. More than that and you can’t give people what they need. The 1:1 cadences break down, the context gets thin, and people feel like numbers.
That logic was built on an assumption: that the manager is the primary source of coordination, context, and feedback for their reports.
That assumption is now optional.
Meta’s new applied AI engineering organization — tasked with accelerating its superintelligence push — is operating with manager-to-employee ratios of up to 1:50. That’s not a thought experiment. It’s double the 25-to-1 ratio that is usually seen as the outer limit of the span-of-control scale. And Meta isn’t alone. According to Gallup, the average number of direct reports per manager rose from 10.9 in 2024 to 12.1 in 2025 — a nearly 50% increase since Gallup first measured this in 2013.
This is directional, not coincidental. The broader “Great Flattening” is already well underway. Gartner projects that one in five businesses will use AI to flatten their organizational structure, slashing over half of current middle management positions. Zuckerberg himself told investors that Meta is “elevating individual contributors and flattening teams” and that projects which once required big teams can now be accomplished by a single, very talented person.
The key insight here isn’t that managers will have 50 reports and somehow run the same playbook at five times the scale. It’s that the playbook itself changes. Imagine an EM whose agents are handling sprint tracking, blocker monitoring, cross-team dependency management, first-pass standards review, onboarding documentation, and status reporting. The human manager’s attention is freed for the things that actually require a human: career conversations, performance calibration, the hard coaching moment, the architectural call that has real consequences. In that world, the bottleneck on span of control isn’t bandwidth — it’s the number of people you can hold genuine context on and advocate for with depth. That number is higher than ten.
The EMs who have figured out how to work with agents — not just tolerate them — will be the ones who survive that flattening with their authority intact. The ones who haven’t will find their role rationalized into someone else’s job description.
What to Do Now
The opportunity here is real, but it requires action before the compression happens to you rather than for you.
Start by auditing your workflow honestly. Map out where your time actually goes. Those 17.9 weekly meeting hours are your starting point. The coordination work, the process overhead, the rituals that exist because someone had to own them — those are your candidates for agent handoff. Not everything will automate cleanly, but most people are surprised by how much will.
Then get back in the design work. Not in a micromanaging way — but in an “I have real opinions and I’m going to develop them” way. Read the RFCs. Push back on designs with technical reasoning. Rebuild the instincts you may have let go soft.
Finally, start thinking about leverage differently. The question isn’t “how many people can I manage” — it’s “how much can I move with the combination of people and agents I have?” That framing will matter more and more as the tools mature.
The best managers have always been the ones who made themselves unnecessary in the right ways. In the AI era, that principle scales further than anyone expected.
The engineering manager role is not disappearing. It’s concentrating. The coordination layer is getting automated, the org structure is getting flatter, and the EMs left standing will be the ones who leaned into technical depth and learned to manage with agents on the team.
That’s a better version of the job. It always was.