Middle management was built for coordination. AI is coming for coordination.

Before you forward this to your manager with a smirk: this is not a "middle managers are useless" piece. The internet has enough of those, and they're mostly written by people who have never tried to get eleven humans to agree on a launch date.

This is a piece about what middle management is actually made of, and what happens when its largest ingredient becomes abundant.

An honest calendar audit

Take a typical manager's week and sort it into buckets. There's coordination: status meetings, syncs, relaying decisions up and down, chasing dependencies, translating between functions. There's judgment: prioritization, tradeoffs, saying no. There's people work: coaching, feedback, hiring, noticing that someone's quietly burning out. And there's politics, which we can debate the legitimacy of some other time.

Every study of managerial time lands on the same embarrassing ratio: coordination eats somewhere between half and two thirds of the week. Not because managers love meetings, but because in a complex organization somebody has to be the router, the node where information from five directions gets synthesized and redistributed. That somebody got a title and a salary band.

The structural truth underneath is uncomfortable. The management layer exists at its current size because coordination was expensive and only humans could do it. Both premises are expiring at the same time.

Coordination is becoming ambient

When people hear "AI is coming for X" they picture a pink slip with a robot's signature. Wrong mental model here. What's actually happening is that coordination is dissolving into the environment where work happens.

Status doesn't need to be collected when an agent sitting in the project space can answer "where are we" on demand, with links to the actual work. Updates don't need to be relayed when an agent notices a decision in one thread and surfaces it to the team it affects. Dependencies don't need to be chased by a human when an agent can watch them and ping the blocking party directly, then escalate to a person only when the ping goes unanswered. Agents are even starting to coordinate with each other: one tracking engineering progress, another tracking the launch checklist, reconciling their pictures of reality without a meeting where two humans read spreadsheets at each other.

The Monday ritual of a manager reconstructing reality from fragments was never the job. It was a workaround for information friction, and the friction is what's leaving.

Meetings won't vanish overnight. They hollow out. First the status portion gets skipped because everyone already saw the summary. Then the meeting shortens. Then someone asks why it exists, and the honest answer is that it holds fifteen minutes of actual decision-making, which was always the point anyway.

The fork in the road

So what happens to a layer whose largest ingredient evaporates? It splits, and the two paths look nothing alike.

Down one path is the judgment-and-people manager. Strip away the routing and what remains is the hard human core: deciding, coaching, developing, resolving conflict, being accountable. This role gets smaller in number and bigger in scope. A manager who isn't spending twenty hours a week as human middleware can support far more output. Managers on this path get measured on decision quality and team growth rather than on crisp status updates. Frankly, this is the job most of them thought they were signing up for.

Down the other path is something newer: the person who designs how work flows through a team that's part human, part agent. What gets delegated where. What a clean handoff between an agent and a person looks like. Where review checkpoints sit, when an agent should call another agent versus stopping to ask a human, what triggers escalation. Coordination architecture rather than coordination labor. The distinction matters. Designing the pipes is a career. Being the pipe was the grind.

The managers most exposed are the ones whose whole value was the routing itself. Always in the loop, superb at status, the person who knows who knows what. That skill set was legitimately valuable, and it's the one being commoditized. If your calendar is mostly transport, the unsentimental advice is to start migrating toward judgment or design now, while it's a choice rather than an emergency.

Two ways organizations will fumble this

The first fumble: delete the layer before the load is gone. Some leadership teams read this trend as "we can cut management" and act early. The coordination work doesn't disappear. It lands, unmanaged, on senior ICs who become managers without the title, the training, or the pay. If you're going to thin a layer, automate the load first and verify it's actually gone. The order of operations is the whole game.

The second fumble is the mirror image: adopt the agents, keep every ritual. The standup still happens, except now everyone reads agent-generated summaries at each other. Coordination cost didn't drop. It grew a subscription fee. The gains only show up when someone with authority kills the rituals the automation made redundant, and killing rituals is itself a management act. Tools don't cancel meetings. Managers do.

There's a quieter trap worth naming too. Coordination meetings were inefficient, but they were where managers accidentally noticed things. The flat tone. The camera off for two weeks straight. The star performer gone quiet. Automate the information flow and you lose the incidental human contact that rode along with it. Good managers will need to deliberately rebuild what used to happen by accident, and the ones who don't will discover attrition problems no dashboard predicted.

Where this leaves you

If you manage people, run the bucket audit on your own calendar and see what percentage is transport. That number is your exposure. Then pick a direction and start collecting evidence: decisions you owned, systems you designed, people you grew.

If you're a senior leader, resist touching the org structure first. Instrument the load instead. How many hours a week does your management layer spend moving information around, and what would have to be true to cut that in half? Restructure after the load drops, not in anticipation.

And if you're an IC eyeing the ladder, notice that the ladder is being rebuilt while you climb. The management jobs that exist in five years will reward judgment and design far more than availability and diligence.

Middle management isn't dying. The router is dying, and for decades the router was welded to the manager so tightly we forgot they were separate things. What's left after the unwelding is the part that was always hardest to hire for: someone with the judgment to decide, the spine to be accountable, and the attention to grow other humans. That job is about to have far fewer meetings. It might finally be a good job.