Why Is Meta Dismantling Its Engineering Teams After Billion‑Dollar AI Bets and Massive Layoffs?

1. Pre‑AI Engineering Culture

For two decades Meta (formerly Facebook) cultivated a fast‑action, rule‑breaking engineering culture, documented in a 70‑page internal booklet called the “Little Red Book” that emphasized speed, fearlessness, ownership, and unconventional thinking.

Engineers enjoyed high autonomy, direct influence from CEO Mark Zuckerberg, and minimal process overhead compared with rivals like Amazon or Google.

2. Shift to AI Investment and Mandatory AI Use

Since 2022 Meta has poured billions into AI, launching the FAIR research group and the GenAI product organization. It released a series of Llama models:

Llama 1 : released Feb 2023, three months after ChatGPT.

Llama 2 : released Jun 2023, developed by the GenAI organization.

Llama 3 : released Apr 2024, the most competitive Meta LLM.

Llama 4 : slated for Apr 2025, described as disappointing.

In June 2023 Meta bought 49% of Scale AI for $14.8 billion, appointing Scale’s CEO Alexandr Wang to lead its AI strategy, and later attempted a $20 billion acquisition of China’s Manus AI, which was halted.

Scale AI provides high‑quality training data, annotation services, and RLHF pipelines that Meta uses to fuel its LLM development.

3. Forced Monitoring and Reassignment

In late April Meta announced a system that records every keyboard stroke and mouse click from all engineers to generate AI training data, with no opt‑out option. After weeks of pushback, a Reuters‑cited internal memo said the company would add a control option allowing a 30‑minute pause per session and possible exemptions.

Simultaneously, 30‑50% of core engineers were reassigned to the Agent Data Optimization (ADO) team for data‑labeling and RLHF work, stripping them of their original project responsibilities.

Meta’s traditional onboarding allowed engineers to choose teams; the new policy forces many senior engineers into low‑value, repetitive labeling tasks, prompting widespread dissatisfaction.

4. Performance Metrics and Token‑Driven Incentives

Meta’s performance review system (PSC) heavily weights impact, code reviews, and lines of code. Since AI became a focus, token generation from AI tools was added as a metric, leading engineers to “spam” AI usage to boost token counts.

The Information reported that Meta engineers consumed 60.2 trillion AI tokens in 30 days, an expense equivalent to $9 billion at Anthropic’s public API rates, even after undisclosed discounts.

5. Security Incident and Fallout

In May 2024 a massive Instagram hack exposed thousands of accounts. Investigation revealed that AI‑generated code changes, combined with a severely understaffed security team (50% turnover after reassignments), allowed a zero‑validation password‑reset vulnerability to be exploited.

Meta’s chief security officer resigned the day after the incident; the CISO Guy Rosen also stepped down.

6. Internal Chaos and Morale Collapse

Wired reported recordings of employee meetings where staff called the environment “a gulag” and expressed feeling like “dogs.”

Chief Product Officer Chris Cox blamed senior leadership for the turmoil, while CTO Andrew Bosworth admitted the AI‑team restructuring was mishandled and promised better communication.

Engineers estimate that 4,500–5,000 of Meta’s ~25,000 engineers are now working on data‑labeling, meaning roughly one in six engineers is diverted from core product work.

7. Broader Industry Reflection

HashiCorp founder Mitchell Hashimoto warned that many companies suffer from an “AI psychosis,” prioritizing rapid AI deployment over reliability, echoing Meta’s situation.

The article concludes that while Meta’s AI bets have driven revenue growth, the aggressive restructuring has turned its engineering organization into a cost center, risking talent loss and long‑term product stability.