Meta Drops AI Metrics From Engineer Reviews
Meta has removed AI tool usage metrics from its software engineer performance reviews after employees began artificially inflating their token consumption to boost their ratings.
Meta is shifting how it evaluates its software engineers by removing artificial intelligence usage metrics from performance reviews. According to an internal memo from executives Maher Saba and Santosh Janardhan first reported by The Information, the company will no longer factor AI dashboards and token counters into employee evaluations. Instead, managers will assess engineers based on the quality, speed, and complexity of their output.
The policy change comes after the previous evaluation system triggered an unintended behavior known internally as tokenmaxxing. To improve their standings on internal leaderboards, some employees began consuming massive amounts of AI tokens without generating real value. This artificial inflation has contributed to a steep rise in operational expenses, with Meta's internal AI usage projected to reach billions of dollars in costs by 2026. To curb these rising expenses, the company plans to introduce strict AI budgets and a centralized tracking dashboard starting in 2027.
Alongside these administrative changes, Meta is currently testing a new autonomous AI agent tool called Hatch. Reported by WIRED, Hatch is designed to execute computer tasks independently on behalf of users. However, the internal rollout has faced resistance from Meta employees. Many workers are hesitant to adopt the tool due to privacy concerns regarding the requirement to link Hatch to their personal accounts.
For software developers and engineering managers, Meta's pivot highlights the pitfalls of using raw activity metrics to measure productivity in the age of generative AI. When organizations incentivize the mere consumption of AI resources, they risk driving up cloud computing bills without achieving corresponding gains in software quality. By returning to traditional performance indicators like code complexity and delivery speed, Meta is signaling that human oversight and efficient resource management remain paramount, even as companies push for deeper integration of automated tools.
This is our own summary of reporting by The Decoder
