Build3 distinct publishers3 min readPublished
Project OT would have cut some teams by 60 percent. Meta's own numbers showed incidents up 40 percent and the time spent on them up 70, and Zuckerberg killed the second wave.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
The load-bearing figure is the ratio between those two code numbers. Internal churn grew roughly six times as fast as work that reached users [11]. That is what it looks like when agent output is measured at the commit, because a commit is not a product.
The repair side compounds it. Growth in response time outran growth in incident count by enough that the average incident consumed about 21 percent more staff hours than a year earlier [14]. So the failures were both more frequent and individually harder to close, and Meta's infrastructure teams tied reliability problems to the surge in AI-generated code [15]. That work lands on engineers reading code they did not write, at volume, under an outage.
Set that against the one case Meta can show working. In April its engineering organization said an internal system compressed roughly 10 hours of performance-investigation work into about 30 minutes and produced code changes for human review [16]. Call it a factor of twenty [17], on a bounded task with a checkable answer and a human at the gate. Project OT tried to extend that model to management, product development and other less structured work before Meta had evidence the technology could carry it [18].
Note what Meta actually contests. It confirmed Project OT and the two-wave structure, confirmed that the most aggressive scenarios contemplated shrinking some teams by up to 60 percent, and said the figure never applied to the whole workforce, that several major units were excluded, and that leaders dropped the second wave before deciding its size [19]. Every one of those points is about scope. None of them is about the productivity data.
The plan also carried a labor bill nobody costed. Meta reassigned some engineers to produce software-engineering tasks used as training data for its coding models [20], and required tracking software on US employees' devices to record keystrokes and mouse activity so agents could learn how people use computers [21]. Employees concluded they were training their replacements and filled internal channels with criticism, according to Reuters as summarised by the-decoder [22]. The same account lists investor complaints about the size of the AI budget among the triggers for the reversal [29], and reports that in July Zuckerberg said the agent technology had not sped up as fast as he expected [24].
What did happen was ordinary. Meta went ahead on May 20 with a reduction affecting about 10 percent of its workforce [6], and between layoffs and transfers some engineering units were down as much as 30 percent by the end of May [26]. The second-quarter filing counted 75,472 employees as of June 30 and said that total still included about 8,000 affected people [27], roughly 11 percent of the headcount on the books [28]. Reuters built its account from internal documents, recordings and interviews with more than 20 people [7]. The cuts were delivered on schedule; the substitution they were supposed to make permanent was not.
Ranked by verification strength, evidence, and original report placement.
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.
A plan code-named Project OT, for Organization Transformation, explored reducing some Meta teams by as much as 60 percent.
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.
Project OT called for two waves: the first scheduled for May, followed by another restructuring in November.
On May 19th, hours before Meta began the first wave, Zuckerberg canceled planning for the November cuts.
Reuters based its account on internal documents, recordings and interviews with more than 20 people familiar with Meta's operations.
Distinct publishers with included, body-backed reporting in this cluster.
mezha.net
1 article · August 26, 2026
runtimewire.com
1 article · August 26, 2026
the-decoder.com
2 articles · August 26, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Meta ran pods-plus-agents for a year and shelved it. Its own scoreboard says why1 distinct publisher
invest
When the marginal bidder is a billionaire, farmland stops being priced off what it grows1 distinct publisher
invest
Meta's Hatch agent tops out at $199.99 a month, with DoorDash and Etsy behind the meter2 distinct publishers
invest
Meta's Pay Structure, Not Its Policy Page, Is What the States Put on the Stand1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Strong documentary base, one primary investigation
The factual core rests on a Reuters investigation built on internal documents, recordings and more than 20 interviews, is partly confirmed on the record by Meta, and is anchored at points to a regulatory filing with specific headcount and spending figures. It is downgraded because all three cluster sources are downstream of that single investigation, the internal productivity and incident metrics are not independently verifiable, and several load-bearing figures appear in only one of the three publishers.
Tooling widely used, org model abandoned
Adoption of the underlying AI coding tooling inside Meta is real and large: internal code changes up 220%, a deployed performance-investigation system, a company-wide device-tracking data program and engineer reassignment to training-data work. Adoption of the thing the story is actually about - reorganizing the company around agent-supervised 'talent-dense' teams - was halted, with the November wave canceled, the tracking program paused and some reassigned engineers returned to prior groups. The split places adoption near the middle rather than high.
Substitution thesis overstated; retreat framing simplified
Positive because the claim being tested - that agents can absorb the daily work of thousands of employees under thin human supervision - is overstated relative to Meta's own numbers, which show activity growth concentrating in internal code and in incident review rather than shipped features. Only a narrow, structured task showed a clean twentyfold gain, and Project OT generalized it before evidence existed. A smaller part of the gap runs the other way in the coverage itself: the shortest treatments compress a multi-factor reversal into 'failing agents' and drop Meta's scoping caveats, while the quantified operational cost is carried by one publisher only.
Company framing, leak motive and outlet framing all in play
Meta has a clear interest in recasting Project OT as scenario planning that was never fully intended, and in protecting a capital-expenditure story that continues after the labor rationale weakened. Employees supplying internal documents and metrics have an interest in stopping cuts and the tracking program. On the publisher side, one outlet's page ends in a subscription pitch built on an anti-hype positioning, and all three build on another organization's investigation, which rewards restatement over verification. These are visible in the supplied material rather than inferred.
High on facts, moderate on single-pipeline risk
Dates, percentages and the filing figures are consistent across the three publishers where they overlap, and Meta confirmed the plan's existence and structure, so confidence in the narrative spine is high. It is held below the high band because every source depends on one investigation, the decisive internal metrics appear in only one cluster source, and there is an unresolved wording difference between 'some teams' and 'many teams' at up to 60%.