In 2023, Mark Zuckerberg gave Meta’s restructuring a name that stuck: the Year of Efficiency. Managers were asked to become individual contributors again or leave. Layers of middle management were cut on the theory that a flatter company would move faster, and that AI tools would soon handle the coordination work that middle managers used to do by hand. Three years later, Meta is quietly asking some of those same people if they would like their old jobs back.
According to Fortune, Meta has begun approaching individual contributors inside its Applied AI division, asking whether they would consider transitioning back into manager roles. The request is described as voluntary. Several of the employees being asked had been managers before, until flattening moved them into individual contributor positions.
The timing is what makes it notable. This is not a company reconsidering an old decision at leisure. It is happening inside the exact division built to prove the original bet was right.
Quick facts
- 2023: Meta’s “Year of Efficiency” flattens management layers, moving many managers into individual contributor roles
- Applied AI Engineering (AAI) launches in 2026 to bridge Meta’s AI research and product execution
- Earlier this year, Meta reassigned roughly 7,000 employees into AI-focused units including AAI, alongside a round of roughly 8,000 job cuts elsewhere
- September 2026: Meta asks individual contributors inside AAI if they would like to return to management, on a voluntary basis
- Some employees being asked had held manager titles before the 2023 flattening
What Applied AI was supposed to prove
Applied AI Engineering was not framed internally as a normal division. It was framed as the place where Meta’s AI research got turned into something that ships, the connective tissue between the teams inventing new models and the teams shoving them into Instagram, WhatsApp and Meta’s ad systems. When roughly 7,000 employees were folded into AI-focused units earlier this year, including AAI, Central Analytics and an internal effort with the deliberately blunt name Agent Transformation Accelerator, the pitch was that this was what a lean, AI-native organization looked like.
It arrived in the same stretch as a round of roughly 8,000 job cuts elsewhere in the company, which made the framing convenient for Meta either way. Employees who kept their jobs were being moved toward the future. Employees who did not were, in the company’s own language, casualties of getting there faster.
The contradiction at the center of this story
Meta spent three years arguing that AI tools would reduce the need for the coordination layer that managers provide. The company is now asking, inside the one division built to prove exactly that thesis, whether some of those managers would like to come back.
Efficiency, then acceleration, then this
| Period | What Meta did | Stated logic |
|---|---|---|
| 2023 | “Year of Efficiency”: managers moved to individual contributor roles, layers cut | Flatter structure moves faster and costs less |
| Early 2026 | ~7,000 staff reassigned into AAI and related AI units; ~8,000 roles cut elsewhere | AI-native teams need engineers close to the model, not management overhead |
| September 2026 | AAI individual contributors asked, voluntarily, to consider returning to management | Not publicly stated; internally framed as reorganization, not reversal |
Why coordination did not disappear
The theory behind flattening was never absurd. A lot of management overhead in a company the size of Meta really is friction: status updates, approval chains, layers of review that exist mostly to justify their own existence. AI tools genuinely do remove some of that, especially in narrow technical workflows where a model can summarize progress or draft a report a manager used to write.
What flattening did not eliminate is the harder kind of coordination: deciding which of several plausible AI projects a team should prioritize, mediating between research and product when their incentives point in different directions, and giving less senior engineers someone to escalate to when a model’s behavior does something nobody anticipated. That work does not show up cleanly in a workflow chart, which may be exactly why it was easy to assume AI could absorb it and hard to notice when it did not.
Meta is far from alone in betting big on AI-native structures. The company reversed course on open-weight models earlier this year too, after arguing internally that open sourcing its frontier work no longer served its interests. A pattern is forming where Meta’s public AI strategy and its internal AI strategy both get revised in the same direction: less certainty than the initial announcement implied.
The pressure underneath the reversal
None of this is happening in a vacuum. The competitive floor in AI keeps rising. ChatGPT crossed a billion weekly users this year even as its market share slipped, a sign of just how large and contested the audience for AI products has become. European rival Mistral raised the continent’s largest tech funding round in September on a pitch built around sovereignty rather than raw model quality, evidence that capital is chasing every plausible angle on the AI race, not just the biggest labs. Meta cannot afford Applied AI to stumble on execution while that landscape moves.
That is likely the real story behind the voluntary offer to rebuild management. Not that Zuckerberg was wrong about AI tools reducing certain kinds of overhead, but that his own company found the limit of that idea faster than the rest of the industry, inside the division with the most to lose if coordination broke down. Whether other companies quietly making the same bet are watching Applied AI as a warning sign has not been reported. It probably should be.
