Project OT considered cuts of up to 60% in some teams; Meta still eliminated about 8,000 jobs in May and redeployed thousands more.
By Ryan Merket · Published
Primary source: Reuters
Why it matters #
Meta tested the strongest version of the AI-labor thesis inside one of the world's largest engineering organizations. Its retreat shows that more machine output can create supervision and repair work instead of eliminating jobs.
Mark Zuckerberg canceled a second wave of Meta layoffs after internal data showed the AI agents underpinning his workforce overhaul were producing far less useful work than executives expected, Reuters reported Wednesday.
The founder and CEO had approved a plan, code-named Project OT for Organization Transformation, that explored reducing some Meta teams by as much as 60%. Meta intended to replace daily work performed by thousands of employees with virtual workers supervised by smaller groups of what an internal planning document called "talent-dense" staff.
The plan took Zuckerberg's long-running campaign for a leaner Meta to its logical endpoint. Between late 2022 and early 2023, Meta cut about 25% of its workforce. In January 2026, Zuckerberg and senior executives gathered at his Hawaii compound and considered a similar or larger restructuring, according to Reuters, which based its account on internal documents, recordings and interviews with more than 20 people familiar with Meta's operations.
Project OT called for two waves. The first was scheduled for May, followed by another restructuring in November. The scenarios combined layoffs, the elimination of open positions, employee redeployments and performance-related departures.
On May 19th, hours before Meta began the first wave, Zuckerberg canceled planning for the November cuts, Reuters reported. Meta proceeded on May 20th with a reduction affecting about 10% of its workforce.
Meta confirmed Project OT and the two-wave structure. Meta also acknowledged that the most aggressive scenarios contemplated shrinking some teams by up to 60%, while stressing that the figure never applied to Meta's entire workforce and that several major units were excluded. Meta said the exercise covered cuts, redeployments and closing vacant roles, and that leaders abandoned the second wave before determining its total size.
The AI output did not match the activity
Meta's internal productivity numbers exposed the central problem. AI tools were helping employees generate substantially more code, yet much of that activity was failing to reach users as finished products.
Code changes to Meta's internal platforms and infrastructure rose 220% from a year earlier, according to an internal post cited by Reuters. Changes that delivered new or improved user-facing features increased 36%. Infrastructure teams had also identified reliability problems connected to the surge in AI-generated code.
Internal posts said major technical and security incidents, including service disruptions and potential data leaks, increased 40% from the previous year. Time spent by employees responding to those incidents rose 70%. The figures suggest Meta's agents shifted work toward reviewing, repairing and supervising machine output, weakening the labor savings used to justify the proposed reorganization.
Meta has documented narrower areas where agents have produced measurable gains. In April, Meta's engineering organization said an internal system could compress roughly 10 hours of performance-investigation work into about 30 minutes and generate code changes for human review. Project OT attempted to extend that model across management, product development and other less structured work before Meta had evidence that the technology could reliably support the shift.
Employees pushed back
The technical problems collided with an employee rebellion over the possibility that Meta workers were being asked to train their replacements. Meta reassigned some engineers to produce software-engineering tasks used as training data for its coding models. Meta also required tracking software on U.S. employees' devices to record keystrokes and mouse activity for teaching agents how people use computers, Reuters reported.
Employees flooded Meta's internal communications system with criticism. Meta's employee sentiment score fell from 74% favorable to 55% in a half-year survey. Between layoffs and transfers, headcount in some engineering units declined by as much as 30% by the end of May.
The first wave remained substantial. Meta's second-quarter regulatory filing listed 75,472 employees as of June 30th and said the total still included about 8,000 people affected by the May reduction, most of whom would leave the reported headcount by the end of September.
After the notifications, Zuckerberg told remaining employees that he did "not expect other company-wide layoffs this year," according to Reuters. Meta d the employee-tracking program, allowed some reassigned engineers to return to their previous groups and increased spending on travel, workplace food and social events.
In early July, Zuckerberg acknowledged at an internal town hall that AI agent development had failed to accelerate as quickly as he anticipated. He said he expected the technology to show greater benefits over the following three to six months.
The retreat leaves Meta spending heavily on the technology that failed to support its most aggressive workforce assumptions. Meta's second-quarter filing showed $31.08 billion in capital expenditures during the three months through June and $349.31 billion in non-cancelable contractual commitments, much of it tied to cloud capacity, servers, networking and data centers.
Project OT established the sequence other employers will now have to defend: Meta planned the smaller workforce first, then tested whether AI could sustain it. Zuckerberg stopped the second round after the productivity data, operational failures and employee response made that order impossible to ignore.