Loading today’s stories
Build1 distinct publisher3 min readPublished
Project OT put individual teams in line for cuts of up to 60% on the assumption that agents would absorb the work. By May the internal productivity read had turned. The second wave never got a total.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
Follow any of these and your For You feed starts watching them — no settings page required.
leadership
Project OT put dates on the layoffs before it put numbers on the agents1 distinct publisher
product
Meta cancelled the AI plan that would have shrunk some teams by 60%1 distinct publisher
product
Meta ran pods-plus-agents for a year and shelved it. Its own scoreboard says why1 distinct publisher
science
Meta counted the code its agents wrote, then cancelled the layoffs1 distinct publisher
The artefact matters more than the percentage. Project OT's scenarios put individual teams down for reductions of as much as 60% while thousands of employees gradually handed parts of their work to virtual assistants, with small groups of specialists left to supervise the result [3][4]. The reporting behind that description is Reuters, summarised by mezha.net [17].
A number like 60% functions as an estimate of residual work, not a headcount decision: how much review, exception handling and cleanup survives an agent run, and how many people that residue needs. Get the residue wrong by a factor of two and you have removed twice the people the work could spare.
Meta ran a pilot that speaks to exactly this. Last July, Ime Archibong's team stood up five small technical groups, two or three engineers and a designer each, working heavily with AI tools [12]. Those groups were meant to drop six-month planning cycles and produce prototypes in four weeks, and an internal guidebook later suggested other units copy the shape [13]. That is roughly fifteen to twenty people across the whole pilot [18], and about a 6.5x compression of the planning cycle if you count six months as twenty-six weeks [19].
Archibong's own framing was a basketball fast break: more shots, taken at better moments, more ideas tested at lower cost and higher quality [14]. It is a fair description of what better tooling does to prototyping. A fast break is counted in shots attempted. For that to transfer into a 60% team cut, the agents also have to carry the work nobody prototypes: on-call, migrations, review of code you did not write, and maintenance on surfaces already shipped. Nothing in the pilot as described measures any of it.
By May the internal read had arrived. Autonomous agents were not delivering the expected productivity gains, staff suspected the AI push was aimed at replacing them, and investors were looking harder at the spend [8]. On the evening of 19 May, hours before wave one, Zuckerberg changed the plan; about 10% of staff went the next day and the November stage was cancelled [7].
More telling than the 10% figure: Meta says the second wave was abandoned before anyone had fixed the total number of employees it would cover [11], which makes the 60% a scenario range rather than a budget line. And Meta's own account of project OT has redeployed workers moving to priority areas including preparing data to train AI models [9]. The AI-first company still wants people, just pointed at the models.
The arithmetic on what was delivered: one internal forecast said the cuts could match or exceed the roughly 25% reduction Meta made in late 2022 and early 2023 [5]. About 10% is 40% of that floor [20]. Meta also pushed back on the 60% reading, saying it described transfers, closed vacancies and cuts to particular teams rather than dismissing 60% of the workforce [10]. That distinction is real, and it is the same distinction that makes the sizing impossible to audit from outside.
My position, in my context: sequence it the other way. Run the pilot, measure the residue on a maintained product with a pager, then size the team. Project OT wrote the sizing first, and the evening of 19 May is what that ordering costs.
Ranked by verification strength, evidence, and original report placement.
Executives considered reducing individual teams by as much as 60%; some employees could be offered new positions and others dismissed.
Meta said the scenarios did include two stages and possible reductions of some units by up to 60%, but stressed that this meant a combination of transfers to other roles, closing vacancies and cuts to individual teams, not dismissing 60% of all staff.
At Meta's annual leadership meeting in January, Mark Zuckerberg and his closest associates discussed a large-scale restructuring of the company in which artificial intelligence systems would perform a significant share of employees' daily tasks.
The internal plan was codenamed "project OT" (organisational transformation) and envisaged turning the owner of Facebook and Instagram into an AI-oriented company.
According to internal planning documents, thousands of employees were to gradually hand over part of their work to virtual assistants, with small groups of specialists supervising those processes.
One internal forecast said the scale of the cuts could equal or exceed the roughly 25% headcount reduction Meta carried out in late 2022 and early 2023.
Distinct publishers with included, body-backed reporting in this cluster.
mezha.net
1 article · August 31, 2026
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.
One relay, one on-record confirmation
Everything a reader sees here passes through two hands: Reuters read the internal planning documents, mezha.net retold what Reuters found. What keeps it above corridor gossip is that Meta answered on the record — conceding the programme, the two stages, even the 60% scenarios — and that the retelling carries concrete, checkable specifics: named executives, dated decisions, team sizes. What is never confirmed is precisely the finding the whole story turns on, the internal read that agents were underperforming. The piece also stops mid-sentence on a device requirement for US staff, so its last thread simply dangles.
The reorg shipped; the agents did not
Two different adoption curves are tangled together in this story. The organisational one is real and traceable: five pilot squads last July, at least 11 units on the model by June, some engineering groups roughly 30% smaller by the end of May, a July model trained on data the redeployed engineers prepared. The technical one stalled — agents were supposed to absorb thousands of people's daily tasks and, by Meta's own internal reading, did not. Adoption of the new org chart outran adoption of the thing that justified it.
The overstatement is Meta's own
The inflated claims in this story do not belong to the publisher; they belong to the internal documents. A forecast that cuts might match a 25% purge produced about 10% in practice — roughly two-fifths of the benchmark. Scenarios of 60% team reductions became, in the company's own retelling, a blend of transfers and closed vacancies. A second wave was cancelled before anyone had worked out its size, which is what an assumption looks like when it never had a number attached. mezha.net is comparatively restrained about all of it, which is why the gap sits moderately above zero rather than high.
Everyone here is positioned
Read the sourcing and the motives line up neatly. Meta confirms the least damaging version — a cost-and-structure programme, 60% meaning reassignment as much as dismissal — and stays silent on the agent data. Zuckerberg explains the April cuts by pointing at capital expenditure, a framing that suits a company defending AI spend to investors who, per the same reporting, were looking harder at it. And the documents reached Reuters from inside a workforce that feared it was being automated away, in a month when even some vice presidents did not know the plan. None of that makes the account wrong; it does mean each detail arrives carrying someone's interest.
Firm spine, soft centre
We can be fairly sure of the skeleton — the programme existed, two waves were planned, one ran at about 10%, the other was scrapped on 19 May — because the company itself concedes most of it. Confidence drops where it matters most: one publisher, one agency behind it, no visible documents, no metrics behind the productivity finding, and no headcount base to turn percentages into people. Enough to act on the direction; not enough to quote the magnitudes.