Meta Platforms Inc has discontinued its practice of evaluating engineers based on AI tool usage, shifting its performance review process back toward a focus on output quality and actual impact

An internal memo distributed to employees outlines the social media giant's move to correct course on its previously aggressive AI adoption tracking system. The policy, which had inadvertently prompted employees to artificially inflate their AI usage numbers, is now being formally abolished

According to the document reviewed by The Information, executives Maher Saba and Santosh Janardhan issued the new performance expectation guidelines. The memo explicitly states that Meta will not use AI adoption dashboards or token consumption data to assess employee performance

Rewriting the Review Framework: Substance Over Metrics

The updated guidance represents a substantial revision to the company's engineering career progression structure. References to specific AI tools and task-specific AI usage instructions have been removed from the previous framework, with new emphasis placed on work quality and actual impact

The memo directs managers to evaluate output quality, execution speed, and the complexity and scope of problems engineers tackle. It also reveals that AI agents currently assist with roughly 93% of Meta's code changes, though the document cautions that AI adoption rates should not be treated as objectives in themselves. The stated goal is better end results: higher-quality deliverables and faster iteration cycles

The Token Windfall: An Unintended Policy Consequence

This policy adjustment comes in response to unexpected outcomes stemming from earlier aggressive AI implementation initiatives. Earlier this year, Meta mandated the use of AI tools within engineering workflows, integrating AI token usage and AI-assisted coding into the hundreds of metrics tracked within its performance management system

In a previous initiative last year, the company gamified AI adoption by monitoring usage patterns and encouraging increased utilization. This approach produced unforeseen behavioral effects, with some employees correlating high token counts with favorable performance ratings. This gave rise to what has been described as tokenmaxxing, a practice where individuals deliberately consume excessive AI tokens to demonstrate their engagement with the tools

Leadership Acknowledged the Issue Months Ago

Meta's management recognized this problem well before the latest directive. Back in April, Chief Technology Officer Andrew Bosworth cautioned in an internal memo that nobody should use AI tools simply for the sake of using them, and that token usage alone does not represent any form of meaningful influence or performance measure

The newly issued memo, co-signed by the two executives, formalizes this position and institutionalizes the shift in evaluation criteria, redirecting performance assessments from behavioral metrics back to measurable work outcomes. A company spokesperson declined to comment on the matter