The sudden unraveling of Meta’s once-celebrated engineering environment has sent ripples through the technology sector, prompting leaders to ask how a company known for bold experimentation could so quickly undermine the very talent that built its products.
In its early years Facebook’s engineering philosophy was encapsulated in the famous ‘little red book,’ a compact manifesto that celebrated speed, fearlessness, and ownership. Slogans like ‘Move Fast and Break Things’ and ‘Done Is Better Than Perfect’ were printed on walls and reinforced through everyday behavior, encouraging engineers to ship code rapidly and learn from failures.
Around the early 2020s Meta began tempering its reckless pace with a new mantra: ‘Move Fast With Stable Infra.’ While the appetite for speed remained, the company started investing in more reliable release pipelines, better monitoring, and a modest increase in testing discipline.
The inflection point came when AI emerged as the next mega-trend. Meta doubled down on large language model research, releasing the Llama series and acquiring a massive stake in Scale AI to secure high-quality training data and reinforcement learning from human feedback pipelines.
One of the first visible changes was the rollout of a surveillance system that logged every keystroke and mouse click of engineers, justified as a means to collect training data for Meta’s new AI models. Employees received no opt-out option, sparking immediate concerns about privacy.
Soon after, a sweeping directive required 30-50 percent of engineers on core product teams to leave their assignments and join the newly created Agent Data Optimisation (ADO) organization, where they would spend full days labeling data, writing test harnesses, and providing feedback on AI-generated code.
Compounding the morale crisis, Meta’s performance review apparatus grew even more punitive. The internal Performance Summary Cycle (PSC) already favored managers who could ‘fight’ for higher rankings by down-rating peers, and engineers learned to game the system by inflating every measurable metric. When token usage from AI tools was added to the scoring rubric, a perverse incentive emerged: engineers began to chase high token counts by over-relying on AI assistants.
The culmination of these pressures manifested in a strikingly embarrassing security incident in late May, when a simple zero-auth password-reset flaw allowed attackers to hijack high-profile Instagram accounts, including that of the Obama White House. Investigations traced the root cause to AI-generated code that had received no human review.
Internally, the response has been a mix of contrition and deflection. Meta’s CTO admitted in an all-hands that the AI reorganization had been ‘atrocious’ and pledged better communication, while senior security leaders resigned shortly after the incident.
Observers beyond Meta have begun to warn that the company’s behavior is not an isolated aberration but a symptom of a broader ‘AI psychosis’ affecting founders who overestimate the autonomy of artificial intelligence.
The talent market is already reacting. Engineers who once considered Meta a crown-jewel destination are now updating their resumes and exploring offers elsewhere, confident that their experience with cutting-edge AI tooling makes them attractive to other Big Tech firms and ambitious startups.
For leaders navigating the AI hype cycle, the Meta case offers concrete lessons: ensure transparency in surveillance, make reassignments voluntary with clear career paths, balance performance metrics with qualitative judgment, and keep a strong human-in-the-loop for critical safety functions.