Meta is adjusting its internal AI performance evaluation methods. The company’s latest memo shows that engineers’ evaluations will refocus on work quality, delivery speed, and real-world impact, rather than the volume of AI tool usage.
Revised evaluation criteria
The memo was jointly issued by Meta executives Maher Saba and Santosh Janardhan. The document clearly states that the company will not evaluate employee impact using AI adoption dashboards or token usage metrics. Previous related statements and task guidelines have also been removed.
Shift from usage to outcomes
The new evaluation criteria emphasize that managers should focus on output quality, work speed, complexity of issues, and scope of coverage. Meta also noted in the document that 93% of code changes are currently assisted by AI agents, but AI adoption rate itself is not the goal.
The practice of "token刷" has been halted.
Meta previously required engineers to use AI tools in their workflows and to include related usage data in performance tracking. As this system progressed, some employees began deliberately increasing token consumption to demonstrate more frequent AI usage. The company later identified this behavior as an issue requiring correction.
The management has previously issued a warning.
In April, Meta's Chief Technology Officer, Andrew Bosworth, stated that employees should not use AI just for the sake of using it, and token usage should not be treated as a metric of influence. This new memo formalizes this stance into performance guidelines.