Meta has shifted about 3,000 engineers — including roughly 70% of new graduates — into an applied AI engineering group focused on building reinforcement-learning tasks and environments.
The push is paired with employee screen, keyboard and mouse tracking, which the report says is meant to capture real workplace workflows that can be turned into higher-value training data and verifiers.
That data strategy follows Meta’s broader AI reset after Llama 4, with aggressive hiring, multihundred-million-dollar to $1 billion-plus pay packages, and a $14.3 billion Scale AI investment tied to recruiting Alexandr Wang and others.
Meta is also racing to expand compute, simultaneously building five 1GW-plus AI clusters and linking campuses over distances of up to 2,000 km, a setup the report says could put its internal AI capacity near OpenAI and Anthropic through 2026-27.
The report argues Meta now has unusual strength in data, talent and compute, but says its current models still trail the leaders and that catching up by year-end is far from guaranteed.