Meta pulls engineers into AI

Meta is moving some of its strongest software engineers into a new Applied Artificial Intelligence engineering group, and Reuters reported the transfers are being assigned across the company rather than filled by volunteers. The change showed up in an internal memo and started reaching selected employees this week. (reuters.com) This is not the team that invents a new model from scratch. This is the team that makes the model usable inside a company the size of Meta, by speeding up tools, evaluations, and the plumbing that turns research into products. (reuters.com) Meta is doing this while it pours money into the rest of its artificial intelligence stack. On January 28, 2026, the company told investors it expects 2026 capital spending of $115 billion to $135 billion, up from $72.22 billion in 2025, with the increase tied to Meta Superintelligence Labs and core infrastructure. (investor.atmeta.com, investor.atmeta.com) That creates a simple internal rule: once a company spends tens of billions on chips and data centers, the next bottleneck is the people who can make those machines produce better answers faster. Moving senior engineers is cheaper than waiting to hire and train a whole new layer of specialists. (investor.atmeta.com, reuters.com) At the same time, Meta has been tightening the culture around how employees use artificial intelligence internally. Fortune reported that the company shut down an employee-built dashboard called “Claudeonomics” after it ranked workers by token consumption and some usage ran into millions of dollars in model-call costs. (fortune.com) A token is one small chunk of text that an artificial intelligence model reads or writes, like counting a conversation by syllables instead of by sentences. If 85,000 employees are all testing models all day, token use turns into a real budget line, not a toy metric. (fortune.com, finance.yahoo.com) The odd part of the leaderboard was that it rewarded volume, not efficiency. Employees could earn labels like “Model Connoisseur” and “Cache Wizard,” which made expensive model use look like a game just as management was trying to industrialize it. (fortune.com, aol.com) Put those two moves together and Meta’s direction is clearer. The company is centralizing the engineers who can make artificial intelligence systems faster and more reliable, while also telling the rest of the company that every token and every test now has to justify its cost. (reuters.com, fortune.com) That is a different phase from the one Meta was in a year ago, when the race was mostly about releasing bigger models and showing benchmark wins. In April 2025, Meta rolled out the Llama 4 family, and by April 2026 Reuters was reporting a company reorganization built around tooling, performance work, and controlled internal usage. (techcrunch.com, reuters.com) For Meta employees, this looks less like a perk and more like a draft. For Meta management, it looks like the moment when artificial intelligence stops being a side experiment inside Facebook’s parent company and starts being run like electricity, with meters, budgets, and the best engineers assigned to keep the grid from wasting power. (reuters.com, fortune.com, investor.atmeta.com)