Meta will remove AI-tool usage from engineer performance reviews, according to an internal memo reported by The Decoder. Executives Maher Saba and Santosh Janardhan said evaluations will instead focus on the quality, speed and complexity of completed work.
The earlier policy turned AI adoption into a measurable review criterion. Employees responded with behavior dubbed “tokenmaxxing,” consuming large quantities of AI tokens to improve the visible metric rather than because the work required them. Meta’s internal AI usage is reportedly on track to cost billions of dollars in 2026.
The company is not abandoning internal AI tools. It plans to introduce budgets and a central usage dashboard in 2027, shifting the emphasis from raw consumption toward cost control and work outcomes. Meta is also testing an autonomous computer-use agent, although some employees reportedly object to connecting it to personal accounts because of privacy concerns.
The reversal demonstrates a familiar measurement problem: once a proxy becomes a target, people optimize the proxy. Counting prompts or tokens can show activity, but it does not establish that software shipped faster, became more reliable or delivered more value.