Meta built and then scrapped an internal plan to cut some teams by as much as 60% and reduce overall headcount by 25% or more by replacing human workflows with AI agents, according to a Reuters report citing scores of internal documents and more than 20 people with knowledge of the effort. The project, codenamed Project OT for 'organization transformation,' was created in January 2026 and called for two rounds of layoffs — the first executed in May, the second canceled shortly after. Meta confirmed the scenario-planning exercise took place and said thousands of employees were redeployed to newly established teams as a result.
Mark Zuckerberg set Project OT in motion earlier this year and directed executives to proceed with changes to management structures. One HR executive told Reuters the scenarios under review would have shrunk Meta's total headcount by roughly a quarter or more. Meta said it canceled the project before finalizing a specific layoff number and that not every scenario was ever expected to move forward.
The savings were earmarked in part for retaining and paying high-performing staff, particularly employees with Llama-adjacent AI engineering skills, according to two people cited in the report. The company also planned to sell AI agents to third parties as part of becoming 'AI native,' a business line Meta launched in June 2026.
Key facts
- 01Meta's Project OT explored reducing some teams by 60% and cutting overall headcount by 25% or more, per an HR executive cited in the report.
- 02The plan called for two rounds of layoffs; the first landed in May 2026, and Mark Zuckerberg canceled the planned November wave.
- 03Internal posts tied AI agents to a 40% year-over-year rise in major technical and security incidents, with employee time spent resolving them up as much as 70%.
- 04Code changes to internal platforms rose 220% year-over-year, but changes that reached users as new or upgraded features rose only 36%.
- 05Project OT was created in January 2026 and followed an October 2025 internal post titled the 'AI-Native Playbook.'
An internal Project OT document defined 'AI native' as a company where AI-ready tools and agents interact, workflows are automated, and new builds are AI-first. The same shift was foreshadowed in an October 2025 internal post called the 'AI-Native Playbook,' which described a pilot to strip out middle management and use agent-assisted analysis to prioritize daily tasks. Meta restructured engineering, research, and at least eight other teams into smaller units under that pilot.
“AI-ready tools and agents interact, workflows are automated, [and] new builds are AI-first.”— Project OT internal document, Meta internal planning document
The productivity math never quite worked. Meta CTO Andrew Bosworth wrote in an early June post that code changes to Meta's internal software platforms and infrastructure were up 220% year-over-year. But changes that actually shipped as new or upgraded features to Meta users were up only 36%. A roughly 6x gap between internal churn and external output is exactly the pattern that makes finance teams question whether agent throughput is translating to product.
The reliability data was worse. Internal posts flagged AI agents making 'large-scale, disruptive actions that humans are unlikely to execute,' tied to a 40% jump in major technical and security incidents year-over-year. Employee time spent resolving those incidents rose by as much as 70%. Meta declined to comment on those specific posts.
In a July company meeting, Zuckerberg acknowledged the pace had slipped. Reuters said it could not determine the single trigger for canceling the November layoff wave, but reports in March and April about impending cuts, along with a separate program that tracked employees' keyboard and mouse input to train AI agents, hurt morale. Meta has since paused the input-tracking program.
Meta also pushed back on one specific framing. The company told Reuters that performance ratings and promotion decisions 'were and are made by people, not AI,' countering internal descriptions of HR employees and AI systems jointly informing promotion calls. The company did not dispute that AI was used in supporting analysis.
Meta is not the only company running this experiment. Salesforce, Klarna, IBM, and Duolingo have all publicly tied hiring freezes or headcount reductions to agent deployments over the past 18 months. What makes the Meta case unusual is the paper trail — a named project, a scenario range of 25% to 60%, a canceled second wave, and internal metrics showing agent activity generating more incidents than shipped features.
The unresolved question is whether the trajectory Zuckerberg described as slower-than-expected is a 2026 problem or a structural one. Agent reliability on long-horizon internal tasks is exactly the capability frontier labs claim they are closing each quarter, and Meta's own numbers suggest that at current model quality, replacing coordinated human teams with agent swarms produces more disruption than throughput.
The Meta reversal is the clearest data point yet that the agent replacement thesis is running ahead of the underlying capability. A 40% increase in major incidents and a 70% jump in remediation time are not rounding errors — they are the kind of numbers that force a CFO to unwind a restructuring mid-flight. For every other CEO drafting an 'AI native' memo, Project OT is now the case study that says the savings model only works if the agents actually hold context across the twenty-step task you handed them, and today's frontier models still do not, consistently, do that at enterprise scale.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
