Why Meta is dismantling its engineering organization

For two decades, Meta ran one of the highest-performing engineering organizations in Silicon Valley. In April 2026, leadership began dismantling it.

The company spent 20 years building a "move fast" culture where engineers chose their own projects, shipped code to billions of users, and operated as a profit center. In the past few weeks, executives reassigned thousands of engineers to data labeling, rolled out keystroke surveillance, tied AI token usage to performance reviews, and gutted security teams. An Instagram account takeover followed—the worst outage in Meta’s history—and the chief security officer and CISO both resigned.

We walked through what happened and why leadership made these decisions.

Meta’s pre-AI engineering culture

Meta’s engineering culture split into two eras.

The first, "move fast and break things," defined the 2010s. Facebook defied conventional best practices and hit a billion users by 2012. The company produced a small book—dubbed the "little red book"—that codified its culture: speed, fearlessness, ownership, thinking outside the box. Mantras like "Done is Better Than Perfect" and "Fail Harder" covered campus walls.

The second era, "move fast with stable infra," emerged by 2022. Recklessness gave way to disciplined shipping. The culture remained engineering-centric: loose processes, minimal documentation, engineers choosing teams during a six-week bootcamp, and internal transfers easy to initiate.

Mark Zuckerberg coded the first version of Facebook himself. He stayed close to engineering and valued software engineers. Engineers felt they worked inside a profit center.

Investing in AI and pressing engineers to always use it

Meta has been the only company among the big five—Apple, Microsoft, Amazon, Google, and itself—without a hardware platform or operating system. After failing to build a mobile OS in the 2010s, Zuckerberg bet billions on VR with Oculus and AR with Meta Glasses. VR didn’t go mainstream.

When AI emerged as a mega-trend in 2022, Zuckerberg moved fast. He assembled the FAIR group and released a series of open-weight models: Llama 1 in Feb 2023, Llama 2 in June 2023, Llama 3 in April 2024. Llama 4, released in April 2025, disappointed.

In June 2025, Meta acquired Scale AI for $14.8 billion and brought Scale CEO Alexandr Wang to take over Meta’s AI strategy. The acquisition of Chinese startup Manus AI for $2 billion stalled after China blocked the deal.

Wang brought expertise in training data, labeling, and RLHF—the reinforcement learning from human feedback flow where people rate AI outputs. This work now falls on Meta’s engineering workforce, and it’s being pulled off with surveillance.

Problem 1: Keystroke and mouse tracking with no opt-out.

In late April, Meta told engineers they were enrolled in a system tracking every keystroke and click to produce AI training data. No way to opt out. The system raises privacy questions: engineers logging into personal bank accounts, writing personal emails, or taking personal calls—all tracked.

Meta held no consultation. Reuters reported in June that Meta dialed back elements of the plan after weeks of pushback, adding controls to pause collection for up to 30 minutes at a time and request exemptions. Engineers in the UK said the system hasn’t rolled out there due to data protection regulation.

Problem 2: 30-50% of engineers on core teams reassigned to data labeling.

Starting in late April, product engineering teams received a mandate: 30-50% of engineers leave their teams and join the ADO org (Agent Data Optimisation).

The word "forceful" matters. From 2004 through last year, Meta gave engineers autonomy. They weren’t hired for a specific team. During a six-week bootcamp, new hires chose a team. Internal transfers were easy. Team selection started dying down around 2024, but any engineer with two years’ tenure knew they once chose their work.

Then they got assigned to a division where impact was unclear, the work was menial, and staying too long would hurt their careers.

Infrastructure and security teams got hit hardest. One engineer described it like The Hunger Games—tributes randomly selected and removed from their environment. Except at Meta, more people were affected: three to five from a 10-person team going from building products used by hundreds of millions to giving human feedback on AI-generated GitHub repos, over and over.

Roughly 6,500 people are in the ADO org—more than at OpenAI or Anthropic. About 4,500 are software engineers. Meta has around 25,000 engineers, meaning one in every five or six may now do data labeling full time.

People are actively looking for new positions. Nobody updates their LinkedIn to "data labeling at Meta." The silver lining: they still have jobs, retained their salary, and weren’t laid off.

Core engineering folks feel treated like trash

Problem 3: A month-long waiting game stoking fear.

On 20 April, Reuters reported Meta planned to lay off 10% of staff in a month. Meta confirmed the news. For four weeks, everyone knew they could be unemployed soon.

Forced reassignments to data labeling started happening simultaneously. Engineers wondered if those moved to data labeling were "safer" than colleagues on product teams. It would be cruel if Meta cut devs reassigned to data labeling.

Problem 4: Performance review is hyper-aggressive, so devs optimize all metrics.

The internal performance review process, PSC (Performance Summary Cycle), is stringent compared to Google and Apple. Managers "fight" over pay packets, "knocking down" engineers on other teams so their direct reports rank higher. Quotas are handed down for workforce splits into buckets. Engineers learn the best way to avoid bad ratings is to have all metrics—impact, code committed, other numbers—higher than peers.

Problem 5: Tokens are measured as part of perf, so devs aggressively optimize for it.

When layoffs were confirmed, engineers learned managers would inspect token count during perf reviews. Low token counts might mark someone as an underperformer and get them dismissed.

The natural reaction: engineers started using AI tools to generate more tokens. Meta had an internal token leaderboard, encouraging tokenmaxxing.

Meta employees used a total of 60.2 trillion AI tokens in 30 days. At Anthropic’s API prices, that would cost $900 million. Meta likely pays a discount, but the total could still exceed $100 million—largely from senseless tokenmaxxing.

The biggest problem: people stop caring about real work and focus on performative work.

Four ingredients, shaken together:

• Tracking keystrokes and mouse clicks of all engineers, where legally possible
• Reassigning a good chunk of engineers to full-time data labeling
• Telling staff 10% of them will be laid off
• Having a culture where devs optimize for any and all metrics measured during PSC
• Measuring token usage as part of PSC

Two outcomes: everyone overuses AI to boost personal stats, and an engineering workforce pretends to work with as much AI and as little human input as possible. It’s a strange incentive where an outage caused by failing to review code properly isn’t grounds for dismissal, but writing code by hand could cost you your job.

Every longer-tenured engineer is seeking a new job, or at least considering it.

Meta has given retention equity packages to several key engineers. These packages make it harder to get matching compensation elsewhere. One engineer who got an equity top-up said the approach helped him decide to leave sooner. He feels bitter about the lack of autonomy and having no control.

Most embarrassing-ever outage

Meta’s core infra and security teams suddenly found themselves severely understaffed. Most folks push AI-generated code merged with AI-only reviews, paying little attention to quality. They deal with the possibility of unemployment while firefighting to operate teams without their best engineers, all with knowledge that AI usage could affect their own job security.

On 30 May, the most embarrassing outage in Meta’s history happened.

Instagram accounts, including high-profile ones like the Obama White House account, got hacked. The attack was simple:

The attacker needs only a username. They use a VPN or proxy close to the victim’s city so Instagram’s security algorithms don’t flag the request. They tell Meta’s support AI the account is hacked and ask it to send verification codes to an arbitrary email address they control. No additional check verifies whether the email belongs to the user. The AI sends the security code to the attacker’s email. The attacker passes it back. The platform hands over a fresh password reset link.

A zero-auth password reset in production.

AI was at the heart of this outage. AI-generated, AI-reviewed code, and gutted security teams together caused the incident:

• Instagram’s Trust and Safety Team lost around 50% of its staff to data labeling and layoffs
• Senior staff got drafted onto AI training tasks
• AI-generated changes with no human input, just another AI code review, were common across the codebase
• The change that caused the outage looked like one of these
• The Trust and Safety team, normally on top of monitoring and alerting, was in disarray

Meta’s Chief Security Officer resigned the next day. The CISO, Guy Rosen, announced his departure. Coincidence? The CISO might have stepped down after warning against the Security org being gutted and being ignored.

The outage resolved on 1 June. Meta shared no postmortem or apology. After a 2021 outage where all Meta services went down for seven hours, the company shared a postmortem and apology. Not this time.

Internal mess

Wired reported on an employee interrupting a livestreamed, employee-only presentation with an expletive-filled outburst about "being the company’s bitch." The individual asked the people leading the call to write to a specific Meta AI executive and "tell him that he’s a piece of shit."

The incident reflected growing frustration inside the Applied AI team, formed in March to support AI researchers at Meta Superintelligence Labs. Three current employees told Wired there was widespread dissatisfaction with how Meta assembled the unit of about 6,500 engineers and product managers and the drudgework they were assigned to improve AI models.

"It’s literally the gulag," one employee said. "You have zero purpose in life all of a sudden, you barely interact with anyone, you just have these tasks every week."

Meta’s Chief Product Officer, Chris Cox, reportedly admitted to staff that Meta’s upper leadership created the mess. During a meeting open to all Instagram employees, Cox addressed the "difficult" and "brutal" environment created by the "insanity of this company" in the past few months.

Cox applauded Instagram employees for launching features and serving around 2 billion users amid what he compared to "running a marathon in the middle of a hailstorm and then, like, your teammate gets replaced and then we’re recording you."

"It’s like what the fuck," he said.

Self-inflicted wounds

Engineers point the finger at two individuals: Mark Zuckerberg and Alexandr Wang.

Zuckerberg has full control over the business. He made the decisions to reallocate engineers to data labeling, roll out tracking software, and lay off 10% of staff when Meta achieved record revenue and profits. The buck stops with him.

But it’s hard to unsee that everything Meta is doing—outside of layoffs—comes from the Scale AI playbook: mandatory keystroke tracking, forced data labeling with 4,500+ engineers, taking best engineers from the heart of the business.

Zuckerberg believes it is more important for Meta to train a coding AI than to operate core products like Instagram, Facebook, or Messenger reliably.

On 12 June, Facebook and Instagram had another SEV0—a full-on outage.

Before all this, Meta was on track to overtake Google as the world’s number one ads business by year’s end. Zuckerberg decided building a coding LLM matters more.

Meta’s leadership is now trying to undo the damage. CTO Andrew Bosworth admitted to staff the AI reorg was atrocious and committed to better communication. Bosworth also said employees would have access to AI coaching tools.

Based on everything learned, Meta’s engineering culture is dead because leadership made clear that engineering is a cost center.

Is it just Meta, or are other companies also acting irrationally?

Mitchell Hashimoto, creator of Ghostty and founder of HashiCorp, said he sees similar behavior by other founders:

"I strongly believe there are entire companies right now under heavy 'AI psychosis' and it’s impossible to have rational conversations about it with them."

Hashimoto lived through the MTBF vs MTTR reckoning of infrastructure during the transition to cloud and cloud automation. Those arguments are rearing their heads again across the software development industry.

"Psychosis folks operate under an almost absolute 'MTTR is all you need' mentality: 'it’s fine to ship bugs because the agents will fix them so quickly and at a scale humans can’t do!'", he said.

"We learned in infrastructure that MTTR is great but you can’t yeet resilient systems entirely. The main issue is I don’t even know how to bring this up to people I know personally, because bringing this topic up leads to immediate dismissals like 'no no, it has full test coverage', or 'bug reports are going down' or something, which just don’t paint the whole picture."

"We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happen so fast that nobody notices the underlying architecture decaying. I worry."

The Instagram takeover outage was exactly this: the engineering team dropped the quality bar for AI-generated and AI-reviewed code, expecting quick recovery from failures. They recovered—after high-profile accounts were hacked and the system was compromised, publicly.

Takeaways

Most of us have something to learn from the disastrous events at Meta caused by hyper-focus on AI to the exclusion of the people who built that company.

In some good news, in the UK, some of the 10% layoffs have been cancelled. At the end of the mandatory consultation period, several infra and security teams learned no engineers on their team would be let go.

Meta has a booming business and already benefits from AI via increased ads revenue. Facebook’s feed is filled with fake, AI-generated videos with hundreds of comments from bots and people who don’t realize it’s AI. Just more content for Meta to show ads next to.

Despite business booming, Meta’s leadership went on a crusade to inflict the most damage possible on its engineering org. Now they’re learning most of it was pointless.

If you’re in a leadership position and feel tempted to make drastic org changes for AI-related reasons, take a deep breath and see where it left Meta.

If you’re an engineer at a company whose leadership over-indexes on AI, consider forwarding this article as additional context.

If you’re hiring standout engineers who are hands-on with AI, it’s never been easier to get talent from Meta. Every engineer there is an early adopter of AI and knows how to build products and AI infra. These folks have soured on the company and its leadership.

Meta’s loss of talent will be the gain of other startups and the rest of Big Tech—one benefit of AI that’s probably unexpected, not least by Meta.

The old mantra of "move fast and break things" has extended to Meta’s engineering org itself, with the company’s rush to over-invest in AI so it avoids missing the latest mega-trend in the tech industry.