Meta Gave Up on a Plan to Replace Workers With AI — Its Own Numbers Told It to Stop
Meta has abandoned a secret internal plan, called Project OT, that would have used AI — software programs designed to do tasks normally handled by people — to replace large portions of its workforce. The company's own data showed that the AI approach was producing fewer results than expected and causing more technical problems. Reuters first reported the story on August 26, 2026, with Engadget providing additional detail the same day.
Project OT, short for "Organization Transformation," ties together two earlier Meta stories: a workforce reduction of about 10 percent that eliminated nearly 8,000 jobs and cancelled 6,000 open roles, and a surveillance program that installed tracking software on US employees' computers to capture mouse movements and keystrokes (Engadget, KRON4).
The plan was laid out at Meta's annual leadership retreat in January, held at Mark Zuckerberg's Hawaii estate. Executives presented a model in which AI programs would handle many daily tasks, overseen by small teams of human employees. Workers would shift into general-purpose "builder" roles, and middle-management layers would shrink. Meta executives were influenced by startups, including some in Asia, that had built their companies around AI from the start (Reuters).
The planned cuts were large. Meta executives considered reducing some teams by as much as 60 percent. One HR executive projected the overall reduction would be at least as large as the company's 2023 cuts, which removed roughly 25 percent of its workforce. Project OT's original form included a second wave of layoffs scheduled for November. Zuckerberg ultimately abandoned, or at least paused, that second wave. Meta acknowledged there had been a second plan but described the more aggressive cuts as merely scenarios under consideration (Reuters).
The decision to pull back lines up with internal data that showed a troubling picture. Meta's own numbers showed that code changes to its AI software platforms and infrastructure were up 220 percent year over year, but new or improved features reaching users rose only 36 percent. Think of it like a factory that produces 220 percent more parts but ships only 36 percent more finished products. Technical and security incidents — meaning bugs, outages, or security problems — rose 40 percent, and the time employees spent fixing those problems grew 70 percent (Engadget).
In July, Zuckerberg admitted at a Meta town hall that he had overestimated how quickly AI would advance. He said the progress of AI agents over the preceding four months had not sped up as expected (Reuters.
The abandoned plan sits at the center of a turbulent period for Meta. The company laid off 11,000 employees, about 13 percent of its workforce, in November 2022 (Guardian). In March 2023, Zuckerberg declared a "Year of Efficiency" focused on flattening organizations, canceling lower-priority projects, and reducing hiring (Meta). By March 2026, Reuters reported that Meta was planning sweeping layoffs as its AI costs mounted, following setbacks with its Llama 4 models, including criticism that Meta had provided misleading information about them (Reuters). Bloomberg reported on May 19, 2026 that Meta had begun job cuts as part of a previously announced restructuring (Bloomberg). In June, Reuters reported that the head of product for Meta's "AI for work" transformation was leaving the company (Reuters).
Meta's AI division has also faced internal strain. The company announced a restructuring of its AI division in August 2025 amid internal tensions (New York Times). Its AI research lab has dealt with a large number of staff departures, following internal disputes over computing power — the processing resources needed to train and run AI systems (The Information). Bloomberg reported in July 2026 that Meta has struggled to turn its AI spending into revenue, though AI has improved its ad targeting (Bloomberg). Additionally, Meta announced it had "right-sized" its Reality Labs investment to ensure sustainability (Meta Developers).
The gap between how much code was written and how many features actually reached users, the rise in incidents, and leadership's own admission about the slow pace of AI progress together explain why Project OT did not move forward. The tracking of employee activity and the scale of proposed cuts, up to 60 percent for some teams, show a company ready to move aggressively on the assumption that AI could take over human workloads at a large scale. Meta's own data became the argument against that idea.
The broader context here is something those of us who covered the cloud buildout and the early mobile transition will recognize: a new technology creates pressure to reorganize around it before the technology itself is mature enough to handle the job. The difference is that those earlier shifts, for all their disruption, did not come with internal numbers showing a 220 percent increase in code changes producing only a 36 percent increase in features reaching users. That gap is the detail worth watching. If other large companies running similar experiments see the same pattern, the push to replace workers with AI will likely slow down further.
What remains is a company still investing heavily in AI, but now without the organizational overhaul that was meant to go with it. Meta's ad-targeting improvements suggest AI is delivering value in specific areas. The bigger question is whether reorganizing a company around AI — collapsing middle management and rebuilding teams around AI programs — can work at the scale of a company with tens of thousands of employees before the underlying AI technology is reliable enough to justify it. On the current evidence, Meta has concluded it cannot. Not yet.
