When a Meta employee hijacked a company-wide livestream last Thursday and screamed at colleagues to tell a senior AI executive he was “a piece of shit,” it wasn’t one person having a bad day.

It was the eruption of months of fury inside one of the most ambitious — and botched — corporate reorganizations in tech history.

Welcome to Meta’s Applied AI unit, where roughly 6,500 engineers are openly calling their new assignment “the gulag.”

The Draft Nobody Volunteered For

Three months ago, Meta pulled off something unusual even by Silicon Valley’s standards: it reassigned thousands of engineers into a brand-new Applied AI division with virtually no warning. Many learned about their new roles through a surprise email. No interview. No discussion. No opt-out.

The mandate was blunt: join or quit.

The reasoning? Meta’s AI models still can’t outperform humans at technical tasks like coding. To close that gap, the company needs real examples of how humans complete everyday computer tasks. And who better to generate that training data than Meta’s own engineers?

In a leaked audio recording, CEO Mark Zuckerberg explained why he preferred drafting employees over hiring outside contractors. Alexandr Wang — founder of data-labeling giant Scale AI, which Meta acquired for $14.3 billion before installing Wang as chief AI officer — knows the contractor world well. But Meta’s engineers have “significantly higher” intelligence than third-party labelers, Zuckerberg said.

It was meant as a compliment. It landed as a sentence.

What ‘Soul-Crushing’ Actually Looks Like

So what are these highly paid engineers actually doing? Generating puzzles. Writing coding problems. Creating training data so AI models can learn how humans use computers.

This is data-labeling work — the exact kind of task that Scale AI built an empire on with armies of gig workers. The difference is that Meta is paying frontier-research salaries for it.

“It’s literally the gulag,” one engineer told Wired. “Most people find the work soul-crushing,” said another.

The organizational structure hasn’t helped. The unit initially had ratios of up to 50 engineers per manager. Workers describe unclear direction, repetitive tasks, no sense of creative contribution, and the nagging feeling they were lured to Meta under one set of promises and handed something entirely different.

CTO Andrew Bosworth has since admitted that leadership did an “atrocious” job explaining the vision. That’s putting it mildly.

The $14.3 Billion Talent-Task Mismatch

Here’s the structural problem Meta doesn’t want to discuss: what it sold to recruits and investors isn’t what it built.

When Meta acquired Scale AI and recruited Alexandr Wang, the pitch was unmistakable — we’re building superintelligence. Wang would run Meta Superintelligence Labs. The name alone screams frontier research. Engineers who joined or stayed did so expecting cutting-edge model work.

What they got was a data pipeline operation dressed up in aspirational language.

This isn’t a vibe issue. It’s a category error in organizational design. Meta paid frontier-research compensation for data-labeling work, and the revolt is what happens when that mismatch becomes impossible to ignore.

If 6,500 engineers are doing puzzle generation, Meta’s de facto approach is “own the data pipeline” — a defensible strategy, but a fundamentally different bet than “build superintelligence.” Investors and recruits were sold one story and got another.

A Wider Morale Crisis

The Applied AI revolt isn’t happening in isolation. Meta’s workplace culture has been deteriorating for months under relentless layoffs, aggressive AI pivots, and increasingly invasive monitoring.

More than 1,600 employees company-wide have signed a petition protesting a program that monitors their clicks and keystrokes — ostensibly for AI training data collection. Not only are you doing work you didn’t sign up for, but the company is watching exactly how you do it.

Chief Product Officer Chris Cox felt compelled to address the “brutal” environment on a call this week. Bosworth promised improvements: better communication, clearer career paths, fewer manager shuffles.

But the fundamental question remains unanswered: will the work itself change?

Zuckerberg’s Memo: Damage Control

On Friday, Zuckerberg sent an internal memo acknowledging that “recent organizational changes had caused distress” and admitting Meta had “made mistakes.” He reportedly promised no further company-wide layoffs in 2026.

The memo reads as crisis communication. It addresses stability without touching the core grievance — the nature of the work. A “no layoffs” pledge doesn’t help when your best engineers are voluntarily walking out the door.

And they will walk. Anthropic, OpenAI, xAI, Mistral, and a constellation of well-funded AI startups are essentially standing job fairs for disgruntled Meta AI talent. The irony is exquisite: Meta’s attempt to accelerate its AI ambitions by conscripting engineers may end up accelerating its competitors’ ambitions instead.

The Lesson for Big Tech

Every major tech company is racing to reorganize around AI. The temptation to treat engineering talent as fungible — to redirect headcount toward whatever the AI strategy demands — is enormous.

But engineers aren’t interchangeable units. A machine learning researcher who joined to work on recommendation systems doesn’t automatically become an enthusiastic data labeler because a reorg email says so. The skills might transfer. The motivation doesn’t.

The deeper lesson is about honesty. If Meta had been upfront — “we need a massive data operation, here’s why it matters, here’s your path back to the work you love” — the reaction might have been different. Instead, it drafted people into a unit with a grandiose name and handed them puzzles.

Silicon Valley has always had a talent for wrapping unglamorous work in aspirational language. But in a labor market where top AI engineers have their pick of employers, that trick doesn’t work anymore.

You can call it Meta Superintelligence Labs. Your engineers will still call it the gulag.

What to Watch

Can Meta fix this without gutting its AI data strategy? The company genuinely needs massive amounts of human-generated training data. That’s not wrong. But the current approach is burning through goodwill at a rate no memo can replenish.

Watch voluntary attrition numbers. Watch where departing Meta AI engineers land. And watch whether the Applied AI unit’s mandate actually changes or just gets rebranded again.

In the race to build superintelligence, your own employees calling the effort “soul-crushing” is not a sign you’re winning.