Meta Stops Using AI to Evaluate Engineers After a Failure

Meta is ending the use of AI usage metrics in the evaluation of engineers and is preparing a budget management system for these usages. At the same time, the Hatch agent is being tested but faces internal reservations related to privacy.

Anticipated Costs and Planned Management by 2027

Internal AI usage at Meta is expected to reach billions of dollars by 2026. To manage these expenses, the company plans to implement specific budgets and a central dashboard starting in 2027. This significant consumption has a notable financial impact. Additionally, trials of the new AI agent Hatch are encountering hesitance internally, with some employees reluctant to connect it to their personal accounts due to privacy concerns.

End of AI Metrics in Engineer Evaluations

AI dashboards and token counters will no longer be used to assess the performance of engineers at Meta. The company will no longer consider the amount of AI tool usage in its evaluation criteria, as announced by Maher Saba and Santosh Janardhan in an internal memo. From now on, the quality, speed, and complexity of work will be prioritized in performance assessments.

Reflection on the Drift of "Tokenmaxxing" and Hatch Tool Testing

The integration of AI usage as an evaluation criterion had led to what some call "tokenmaxxing," where employees heavily utilized AI tokens to improve their standings in internal rankings. Meta is also testing Hatch, an AI agent touted as capable of autonomously managing computing tasks. The policy change was formalized in an internal memo, the content of which has been reviewed.