Meta halted Project OT’s AI workforce plan

Project OT was Meta’s reported plan to reorganize daily work around AI agents and smaller human teams. The original planning process is no longer happening, and Mark Zuckerberg canceled the initiative’s planned November phase on May 19, 2026. Meta still carried out about 8,000 layoffs—roughly 10% of its workforce—the following day and moved about 7,000 employees into priority AI-related work.

That is the clean answer. The more important one is what it does not mean: the available evidence does not show that Meta abandoned its broader AI strategy. It shows that the most aggressive version of the workforce redesign did not proceed as planned.

Project OT was a plan for smaller human teams and more AI agents

Project OT stands for Organization Transformation. Its stated direction was to make Meta “AI native”: AI-ready tools and agents would interact, workflows would be automated, and new products would be built with AI at the center.

In practical terms, the model described in the reporting would have placed more responsibility for routine work on software agents while leaving smaller groups of specialists to guide, review, and manage that work. Product teams that previously included roughly 10 to 20 specialists were described as potentially shrinking to pods of three to five people. Some managers would also oversee larger groups as middle-management layers were reduced.

The headline numbers surrounding the plan need careful handling. The most dramatic reduction scenarios concerned particular teams, not a confirmed cut to Meta’s entire workforce. A separate estimate attributed to an unnamed HR executive described a possible overall reduction of about 25% or more, but that was a reported scenario rather than an implemented Meta-wide target.

Meta carried out one phase and canceled the next

Project OT was organized around a May phase and a broader November phase. The November stage was canceled on May 19, hours before the first layoff wave. The following day, Meta carried out approximately 8,000 layoffs, or about 10% of its workforce, while moving thousands of other employees into priority AI work.

Those events are easy to blend together, but they are not the same thing. The layoffs and redeployments happened. The broader November phase did not. Meta acknowledged the exercise and said it did not move forward with every scenario.

The evidence does not establish exactly why Zuckerberg canceled the November phase. Technical concerns, employee backlash, morale problems, and the difficulty of making AI agents reliable all formed part of the surrounding story, but none can be identified as the decisive cause.

The difficult part was not generating work—it was owning the result

AI agents can produce drafts, code, recommendations, and other intermediate work. An organization still needs people to check accuracy, handle exceptions, make judgment calls, and accept responsibility when the output causes trouble. That distinction sits at the heart of Project OT’s reported difficulties.

The initiative was associated with concerns about reliability in AI-generated code and warnings that agents could take disruptive actions humans would not approve. The practical challenge was therefore larger than asking whether an AI system could complete an isolated task. Meta had to determine whether smaller teams could safely supervise systems operating across complex products and infrastructure.

That is a much higher bar. A faster output pipeline is not automatically a more productive company if the review, repair, and risk-management work expands along with it.

Employee monitoring added another layer of tension

The proposed transformation also collided with workplace concerns. Meta paused a program that captured employee keystrokes and mouse movements for an AI-training effort after internal backlash and privacy concerns. The episode made the human side of the plan impossible to ignore: an initiative sold around automation and efficiency was also changing how employees were observed and evaluated.

Reporting also described a possible role for AI systems in management support. Meta rejected the idea that AI made performance ratings or promotion decisions, stating that those decisions were made by people, not AI.

That distinction matters. Using software to organize information for a manager is not the same as handing the software authority to decide who gets promoted. The two ideas are often mashed together in AI workplace discussions, even though they carry very different consequences.

Project OT ended; Meta’s AI direction did not necessarily end

The original Project OT planning process is no longer happening, and the November phase was canceled. Those are the clearest conclusions available from the reported timeline.

The broader picture is different. Mark Zuckerberg later acknowledged that AI-agent development had progressed more slowly than expected, while continuing to support Meta’s larger AI direction. That suggests a change in pace or execution—not proof that Meta had abandoned AI-native work altogether.

It also leaves an important question open: whether the same organizational ambitions might return in a different structure or under a different name. The evidence does not establish that a successor plan exists, so it would be premature to treat one as fact.

For readers tracking the future of work, the useful lesson is less dramatic than “AI replaced Meta’s employees” and more revealing: AI-generated activity is not the same as completed, reliable, valuable work. Project OT tried to push that distinction into the structure of an entire company. Meta halted its next major phase, but its continued investment in AI means the experiment’s central question has not gone away.

Bottom line: Meta stopped the original Project OT planning process and canceled its November phase after implementing a separate May round of layoffs and redeployments. The evidence supports a narrower conclusion than a total retreat: Meta’s aggressive workforce redesign was halted, while its broader AI strategy continued.