Inside Meta's Aborted Plan to Replace Thousands of Workers With AI

For thousands of Meta employees, the past year has felt like living inside an experiment nobody agreed to join. The company drew up plans, internally called Project OT, to shrink teams across the business by as much as 60 percent in two waves, betting that artificial intelligence could do the work of the people it let go. It didn't happen as planned. Staff pushed back, the AI agents underperformed, and hours before a major layoff in May, CEO Mark Zuckerberg called off further rounds of cuts.

That reversal matters far beyond Menlo Park. Reuters technology correspondent Katie Paul, who broke the story, fielded audience questions in an AMA about what really went wrong. Her answers paint a picture of a company that misjudged both its technology and its people, and one whose stumble may still shape how other employers approach AI-driven restructuring.

Think of Project OT as a factory floor experiment gone sideways. Executives assumed AI tools could slot into roles once held by humans the way a new machine replaces a manual task on an assembly line. But software that writes code or manages workflows doesn't behave like a predictable machine. It behaves more like a new hire with no institutional memory, prone to unpredictable mistakes at scale. That mismatch between expectation and reality is where things began to unravel.

What Went Wrong Inside Meta

Zuckerberg himself has acknowledged missteps. In an internal Q&A, he admitted the company got the timing wrong. Chief Technology Officer Andrew Bosworth, known internally as Boz, was blunter still, telling colleagues in a comment thread that Meta did an "atrocious" job explaining the vision to its own workforce. That kind of internal admission, from executives at that level, is rare and telling.

The technical failures were just as serious. Site reliability emergencies, called SEVs inside Meta, rose 40 percent compared to the previous year. AI-related incidents proved harder to predict than human-caused ones, and staff spent 70 percent more time firefighting them. Engineers found themselves trying to build AI tools to catch bugs created by other AI tools, a problem Paul notes remains unsolved not just at Meta but across the industry.

Money pressures loomed over all of it. Meta is grappling with a cash crunch tied to its enormous spending on AI infrastructure, the kind of capital outlay measured in billions of dollars for data centers and computing power. Zuckerberg told employees in late April that the company essentially has two cost centers: compute and people. Spend more on one, he said, and there's less to spend on the other. Paul stopped short of calling the layoffs "desperate," but the arithmetic behind them was plain enough.

There's also a human toll that's harder to quantify. A lawsuit alleges Meta used AI to target workers with medical conditions during the layoffs, and similar allegations have surfaced regarding the company's performance-review system, which grew notably aggressive last year. Paul noted that most of these disputes get funneled into private arbitration, shielded from public view by employment contracts. That makes the full scope of the harm nearly impossible to track from the outside, even as the allegations point to a pattern worth far more scrutiny than it's gotten.

Not everyone at Meta stayed silent. Paul was careful not to reveal sources, but she said some employees believed both their colleagues and the wider public deserved to know what the company was attempting, especially given how easily this kind of overhaul could become a template other corporations quietly adopt. Around the same time, workers organized in protest of Meta's push to capture employee mouse movements and keystrokes for AI training data, a move that drew scrutiny well beyond the company's walls.

Meta hasn't abandoned AI. It has walked back the more aggressive push to use it everywhere for its own sake. Internally, the phrase making rounds is "tokenmaxxing," industry slang for cramming AI into every workflow regardless of whether it actually helps. That pressure is reportedly easing. The new guidance, according to Paul, is to use AI when it makes sense, not simply because leadership wants adoption numbers to look good.

Why It Matters

What happened at Meta is a warning sign worth heeding well beyond one company's org chart. When a firm with Meta's engineering talent and financial resources struggles to make AI agents reliably replace skilled workers, that tells you something about where the technology actually stands, not where marketing decks claim it stands. The gap between promise and performance was wide enough to cost the company real money in firefighting hours, site outages, and lost productivity.

For workers everywhere, the episode offers a measure of reassurance mixed with genuine caution. Reassurance, because it shows automation ambitions can collide hard with technical and organizational reality, and companies can and do walk back plans when the numbers don't add up. Caution, because the underlying pressure that drove Project OT in the first place, the need to fund massive AI infrastructure spending somehow, hasn't gone away. Companies squeezed between compute costs and payroll will keep looking for ways to shrink one side of that ledger.

The unresolved questions about disability and medical leave discrimination in the layoffs deserve particular attention. If AI systems are being used to identify workers for cuts in ways that disproportionately affect people on protected leave, that's not just a labor story. It's a legal and ethical one, with implications for how companies deploy algorithmic decision-making in employment decisions more broadly. Arbitration clauses may keep individual cases out of public courtrooms, but the pattern itself is a matter of public interest, and one likely to draw more attention as similar restructuring efforts unfold at other firms watching closely to see what Meta learned, and what it still hasn't figured out.