Meta Bet AI Would Replace Middle Managers; Its AI Division Needs Middle Managers

Earlier in 2026, Meta reassigned roughly 7,000 employees into a new division called Applied AI, built to bridge the gap between the company’s research labs and the products it actually ships. As part of that reassignment, some workers who had previously held manager titles were moved into individual-contributor roles, in keeping with a company-wide push toward flatter teams — up to fifty individual contributors reporting to a single manager, a ratio Mark Zuckerberg had championed as central to what he called the “year of efficiency.” That same push coincided with Meta cutting about 8,000 jobs, roughly 10% of its workforce, in May, with the layoffs falling disproportionately on managers. Fewer bosses, leaner teams, and AI doing more of the coordination work that middle management used to handle: that was the pitch.

As of this week, Meta is voluntarily inviting individual contributors inside Applied AI to consider becoming managers again.

The company is careful to frame this as optional and narrow — a program limited to one division, not a company-wide reversal of the flattening strategy. But read the plain sentence next to the plan it’s reversing and the shape of the story doesn’t need much editorializing. Meta un-managed a group of employees on the theory that a flatter structure, aided by AI, would run just fine without them. Then it built the actual division tasked with deploying that AI, and discovered that division needed managers after all. The bet paid off everywhere except in the one place built to prove it.

One outlet summarized the situation about as plainly as it can be summarized: the shift “represents a partial reversal of Meta’s recent push toward flatter teams and fewer middle managers as the company seeks greater efficiency while increasing its investment in artificial intelligence.” Two clauses, same sentence, opposite directions — cut management to get efficient, then add management back, in the AI division specifically, in pursuit of the same efficiency.

It’s worth being precise about what actually happened to the people in the middle of this. Several of the roughly 7,000 employees moved into Applied AI did not ask to be reassigned, and some pushed back hard enough that Meta let them leave the division rather than stay in a role they hadn’t chosen. That’s the backdrop against which the new invitation lands: having involuntarily reclassified some managers as individual contributors months ago, Meta is now asking, voluntarily this time, whether anyone would like their old job back. It is, structurally, a rehiring process for a role the company eliminated on purpose, aimed at people who may still remember exactly why it was eliminated.

Zuckerberg’s own framing of the original push was that AI would help teams “speed up product development” — the implicit promise being that software could absorb enough of the coordination overhead that a human layer of it became optional. What Meta appears to have run into instead is that building and deploying frontier AI at scale is itself an unusually coordination-heavy activity: model training runs, safety reviews, infrastructure procurement, and product integration across a division of thousands of people don’t parallelize themselves just because the product being built happens to be intelligent. The flatter structure that might work for a stable, mature product team turns out to strain considerably harder when the team is simultaneously trying to ship the technology that was supposed to make the structure unnecessary.

None of this required anyone at Meta to say the word “irony” out loud, so nobody did. The company’s public language stays firmly in the register of “balancing leaner teams with the need for additional leadership” — a phrase that manages to describe a reversal without conceding that anything was reversed. But the sequence of events speaks for itself: flatten the org on the theory AI reduces the need for managers, assign the people building that AI to a newly flattened structure, watch that specific structure buckle first, and quietly ask for management back exactly there. If AI was going to prove it could run a company with fewer humans in charge, the AI division was supposed to be the strongest evidence. Instead it’s the first place Meta has walked the experiment back.

The program remains voluntary, and Meta has said nothing about extending it beyond Applied AI. Whether that holds probably depends on how the rest of the company’s flattened divisions are doing — and on whether “efficiency,” a year in, still means what it meant in January.

Sources: