Meta tried to turn its own workforce into an AI-first experiment. The plan ran into the oldest problem in business: output is not the same as useful work.
Mark Zuckerberg did not just talk about AI changing work this year. According to Reuters, he and Meta's senior team built an internal restructuring plan around it, then stopped the most aggressive part before it could run its full course. That is the story. Not that AI agents are useless. Not that Meta is done spending. The real point is sharper: one of the richest companies in technology tried to make the productivity math work inside its own walls, and the numbers did not behave.
The plan was code-named Project OT, short for Organization Transformation. Reuters reported on Aug. 26 that Zuckerberg and other Meta leaders discussed it in January at the company's annual leadership retreat at his Hawaii compound. Internal planning material described an "AI native" future in which AI-ready tools and agents interacted, workflows were automated and new work was built around AI from the start. The human side of that plan was just as blunt. Some teams could shrink by as much as 60 percent, not across the whole company, but in the most aggressive scenarios for certain units.
Meta confirmed to Reuters that Project OT existed and that the exercise looked at redeployments, closing open roles and cuts. It also said the company never assumed every scenario would go ahead and never planned to cut 60 percent of its total workforce. That distinction matters. A 60 percent company-wide layoff would be a different story. Still, if you're an employee sitting inside one of the teams being modeled, a scenario is not a harmless spreadsheet. It is your job with a probability attached.
It didn't work cleanly.
Zuckerberg Personally Lobbied Trump to Kill a National AI Regulator Plan Mark Zuckerberg personally lobbied President Trump against a proposed FINRA-style national AI regulator during a private call the week of August 17, 2026. The plan, championed by Google DeepMind's Demis Hassabis, remains under White House consideration despite Zuckerberg's objections, marking the second time this year he's helped reshape federal... - how Zuckerberg lobbied Trump to block AI regulator - Zuckerberg's call with Trump about national AI regulation
Meta still carried out the first round. Reuters reported in May that the company planned to lay off about 10 percent of its workforce, close to 8,000 employees, and move 7,000 workers into new AI workflow initiatives. Bloomberg also reported that Meta would leave 6,000 open roles unfilled. Those are not small internal tweaks. They are the kind of moves a company makes when it believes the org chart has to change before the product roadmap can.
The Code Went Up Before The Product Did #
The most useful number in the whole story is not the layoff figure. It is the gap between activity and delivery. Reuters reported that an early June post by Meta CTO Andrew Bosworth said code changes to internal software platforms and infrastructure were up 220 percent year over year. Changes that reached users as new or improved features were up 36 percent.
The gap matters.
If an AI agent helps produce more code but does not help ship more things users can see, you have not proved productivity. You have proved motion. Anyone who has worked near software knows the difference. More code can mean a better product. It can also mean more review, more bugs, more cleanup and more meetings about why the system did something nobody asked it to do. Reuters reported a darker version of that problem too. Internal posts tied unchecked AI agent activity to "large-scale, disruptive actions" that humans were unlikely to take, and major technical and security incidents rose 40 percent from a year earlier. Employee time spent resolving those problems rose as much as 70 percent. That is not a gain. That is a new workload dressed up as automation.
Staff pushed back as well. Reuters described employees protesting the changes with flyers, internal posts and a petition over mouse-tracking software meant to help train AI systems to imitate how people use computers. Business Insider reported on Sept. 11 that Meta later asked some employees in its Applied AI division to return to management roles, a reversal of the company's earlier drive toward flatter teams. Look at the timing. First came the push to remove layers. Then came the request for some of those layers to come back.
Meta's Spending Makes The Lesson Harder To Ignore #
Zuckerberg has not backed away from AI. Fortune reported that Meta raised its 2026 capital expenditure guidance to as much as $145 billion after first-quarter results, with the company citing higher component prices and more data center costs. The Motley Fool's transcript of Meta's second-quarter call later showed the range narrowed to $130 billion to $145 billion. This is still one of the largest AI infrastructure bets on the market.
Zuckerberg bets billions of personal AI agents on Meta's most expensive quarter ever Meta CEO Mark Zuckerberg predicted billions of personal AI agents within five years on the same Q2 2026 earnings call where profit fell 14%, free cash flow dropped 91%, and the stock tumbled nearly 10% after hours. The gap between the vision and the current financials is the real story. - personal AI agents for consumers - Meta expensive quarter cash flow concerns
Investors are right to ask what they are buying. Fortune quoted Melissa Otto, head of Visible Alpha Research at S&P Global, saying Meta's higher capex raised the question of the real return on all that spending. That is not a hostile question. It is the only serious one. If the money is going into chips, data centers and expensive AI talent, the return has to show up somewhere clearer than a dashboard of generated code.
Zuckerberg has already conceded part of the problem. In a July internal town hall, Reuters reported that he said the trajectory of agentic development over the prior four months had not accelerated the way Meta expected. He also said the reorganization had not been as clean as it could have been and that he expected more meaningful benefits within three to six months.
Frankly, that should travel beyond Meta.
If you're running a smaller company and thinking about replacing whole chunks of work with agents, Meta's retreat is not a reason to ignore AI. Don't bother drawing that comforting lesson. The better lesson is harder: measure delivered work, not machine activity. Count shipped features, resolved tickets, customer outcomes and incident load. Count the cleanup too. If you leave out the cleanup, you're only measuring the pretty half of the experiment. Meta has more compute, more researchers and more cash than almost any company trying this. It still couldn't make the layers disappear. The management layer, the review process, the dull human judgment around work - none of it vanished just because an agent could write code. Business leaders should sit with that before they turn their own staff into a cost-saving model.
Meta declined to comment to Reuters on several details of Project OT, and Reuters said it could not determine exactly what prompted Zuckerberg to change course. That is a useful limit on the story. The company has not abandoned AI, and it has not published a final verdict on whether feature output will catch up with code volume. For now, the cleanest fact is the one Meta's own sequence gives you: the first wave went ahead, the second wave did not.
Also read: Anthropic Is Reportedly Heading Toward a $2 Trillion Nasdaq IPO in October • An AI Chatbot Can Now Unmask Secret Ballots in Georgia's Voting System • A Solo Developer Used GPT-6 Astra to Code a Finished Playable Game
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.