Meta has told its engineers that how much they use AI will no longer decide how they are rated. New performance review guidance issued this week drops the language that tied evaluations to AI usage and to the internal AI Native designation, replacing it with a softer line saying the work "can be supported by AI or other means." Three employees who received the message described the change to Wired. The Information reports that engineers across the company were told management will not use AI adoption dashboards or token counts to judge impact. That is a reversal of the system Meta ran for most of the past year, when workers were graded on their AI-driven impact and sorted into tiers named AI Native, AI First and AI Enabled. The result inside the company was “tokenmaxxing,” a race to consume as many AI tokens as possible so the numbers on the dashboard looked right. For engineers who spent months prompting chatbots to protect their ratings, the memo lands as a quiet admission that the metric was never the point.
The AI-driven impact standard arrived almost a year ago and quickly turned into a competition. Internal dashboards tracked token consumption, and one employee built a leaderboard that ranked colleagues by how much AI they were burning through, handing out labels to the heaviest users. It was pulled in April after details leaked. Some staff made so many AI agents that others built agents to find agents and agents to rate agents. The pressure had consequences beyond bruised morale. About two dozen employees sued in July, arguing Meta broke US antidiscrimination law when it laid them off in May. Workers on health and family leave say they could not accumulate usage and were marked down for it. Meta has denied the allegations and the case is ongoing.
What managers are being asked to do now is look at what an engineer produced rather than how they produced it. The guidance restores impact as the measure and treats AI as one route to it. Meta spokesperson Tracy Clayton told Wired the update simply emphasises what was always true, that employees are assessed on their contributions, and said the AI Native label was never used in evaluations. Employees read it differently. Several told Wired the wording change is subtle but welcome, because it removes the pull to use AI in situations where it does not help. Others are not convinced the expectation has gone anywhere. Leadership still wants proof that engineers can wield these tools well, and token consumption is still climbing as staff test the company's newest agentic project, Hatch.
The push was never only about reviews. Meta ran AI Transformation Weeks in March, told designers to try coding and coders to try design, and later rationed AI usage after supply pressure forced a rethink. It also installed software on US employees' machines to capture keystrokes, clicks and screen content for model training. Over 1,000 workers signed a petition against it and the programme was paused. Internal numbers gave the strategy its own problem. Code changes to internal systems jumped 220 percent year over year, while changes that reached users rose 36 percent. The volume was real. The impact was not. Dropping token counts from reviews is Meta conceding the same thing on paper.
