The Line Nobody Expected to Hear Live

“Tell him he’s a piece of sh*t.”

That line was said live. During an internal presentation streamed to thousands of Meta employees. By an anonymous employee who hijacked the call to express what many felt but nobody said publicly. The target: one of Mark Zuckerberg’s top AI executives.

One of the presenters covered their face with their hands.

When I read the Wired investigative report (published June 12, 2026) and the subsequent coverage from TechCrunch, Raw Story, TechTimes, and dozens of other outlets, my reaction was a mix of shock and a thought I can’t ignore: this was predictable.

Because when you transfer 6,500 of the planet’s most talented engineers to write “coding puzzles” to train AI — with no opt-out — and then monitor every click, keystroke, and mouse movement, the question isn’t if the revolt will happen. It’s when.

What Happened (The Timeline)

March 2026. Zuckerberg decides Meta is going “all in” on AI. To accelerate, leadership creates a new unit called Applied AI Engineering under Maher Saba, a 12-year veteran who previously led Reality Labs — the division that burned through $83 billion on the metaverse.

6,500 engineers and product managers receive a short email: either accept the transfer to the new AI division, or quit. No negotiation. No consultation. Employees start calling themselves “draftees.”

The work: write 2 coding puzzles per week to serve as training data for Meta’s AI models. Engineers who built the ad auction system, messaging infrastructure, and content ranking algorithms — now generating mundane tests.

May 2026. Meta implements the Model Capability Initiative — a monitoring program tracking clicks, keystrokes, mouse movements, and screenshots on work machines. No opt-out. Stated purpose: capture “how smart people use computers” to train AI. Felt purpose: surveillance.

June 2026. Over 1,600 employees sign an internal petition demanding an end to monitoring. Flyers appear posted in company bathrooms. UK staff join the UTAW union (United Tech and Allied Workers). The division earns the internal nickname “The Employee Data Extraction Factory.”

June 12, 2026. Wired publishes the investigation. The “gulag” quote goes viral.

June 13, 2026. Zuckerberg writes an internal memo admitting the changes “caused distress” and that “given the complexity of these changes, we’ve made mistakes and will almost certainly make more.” He ruled out further mass layoffs for the rest of 2026. Chris Cox, Meta’s CPO, called the environment “brutal.”

The Existential Paradox

This is the part that kept me up at night.

The Applied AI engineers know what they’re doing. They’re too smart not to see it: they’re being forced to use their human intelligence to train the exact algorithm that will eventually be used to replace them.

As one employee told Wired: “You suddenly feel like you have zero purpose in life. You spend the entire week just generating dull tests.”

These aren’t interns. They’re engineers earning $300,000-$500,000/year. Who built products used by 3 billion people. And now they write coding puzzles as if they were Mechanical Turk crowdworkers — but with more invasive corporate monitoring.

The reason Meta needs them, not external contractors, is revealing: according to leaked Zuckerberg audio, Meta’s AI models “still lacked the knowledge to outperform humans at technical tasks like coding.” The company needs the tacit knowledge of its best engineers — the kind of expertise I discussed in the post about the end of the infinite internet. The “new gold” that no public dataset contains.

The problem? They’re extracting that gold from people who hate the extraction.

Leadership’s Response: Pizza and a Hackathon

Facing a workforce on the brink of mass resignation, Zuckerberg’s response was to announce a global AI hackathon with prizes and… free pizza.

The attempt to defuse a deep cultural crisis with superficial perks generated mockery in internal forums — especially when employees recall that Meta generates about $56 billion per quarter.

As someone on Pragmatic Engineer (Gergely Orosz’s newsletter) summarized: Meta’s management so far has consisted of starting with a ratio of 50 engineers per manager, zero context, zero autonomy, and then trying to fix it with pizza.

The Connection to Everything I Write

When I read this story, I saw echoes of at least five themes I explore on this blog:

The End of the Infinite Internet. Meta needs high-quality training data that doesn’t exist on the public internet. The solution? Force their own engineers to produce it. It’s the most brutal possible confirmation that human expertise is the new scarce resource — and that some companies prefer to extract by coercion rather than pay for it with dignity.

The Hidden Price of “Free.” If cognitive offloading atrophies user brains, what happens to engineers forced to produce training data on tasks far below their capability? Professional atrophy is the price these 6,500 are paying.

The Trust Paradox. If 76% of Americans don’t trust AI, stories like this don’t help. AI trained by frustrated engineers under surveillance, producing data by obligation — will hardly achieve the quality the market expects.

The Pope’s Encyclical. Leo XIV denounced “a new form of slavery disguised by modern corporate branding.” Meta’s engineers could have written that sentence.

95% of Projects Fail. BCG’s 10/20/70 rule says 70% of success comes from people and change management. What did Meta do? Ignored the people. Monitors every keystroke. And acts surprised when revolt erupts.

What I Really Think

Meta has the right to reallocate employees. That’s corporate management. Companies pivot. Priorities change. Nothing abnormal there.

What’s indefensible is how they did it. Short email with no option. Repetitive work far below qualification. Keystroke and click monitoring with no opt-out. Zero consultation. Zero autonomy. And when revolt arrives, pizza.

AI trained by frustrated, surveilled engineers will hardly achieve the brilliance the market expects. Training data quality is only as good as the morale of those producing it. And morale in Meta’s Applied AI is, in its employees’ own words, “soul-crushing.”

The final irony: Meta invested $15 billion in Scale AI (which I discussed in the human expertise post) — a company specializing in recruiting and orchestrating high-quality human judgment with dignity and aligned incentives. And simultaneously, it treats its own engineers as the internal, coercive version of what Scale AI does externally with respect.

Conclusion: The Human Limit of the Technology Race

Meta’s crisis is a warning for the entire market. In the rush to win the AI race and deliver results to shareholders, Big Tech companies are treating their greatest talent as data-labeling machines.

AI accelerators are fueled by high-quality human data. But if companies crush the dignity and purpose of the professionals generating that data, it can backfire.

You can force an engineer to write puzzles. You can’t force them to care. And data generated without care trains models without excellence.

Share if this resonated:

6,500 of the world’s best engineers, writing puzzles under surveillance, calling the work a “gulag.” If this is the cost of the AI race, maybe it’s time to ask: a race to where?

Read Also