Meta announced adjustments to its AI performance evaluation standards, explicitly condemning "token-swiping" behavior.

Meta has officially stopped using AI usage to measure engineer performance, and is now returning to evaluating work quality and actual impact.

A recent internal memo from Meta reveals that the social media giant is correcting its previously aggressive AI usage metrics mechanism—a "token-farming" trend fostered by the company's own policies is being explicitly halted.

On September 2nd, according to an internal memo obtained and reviewed by The Information, Meta executives Maher Saba and Santosh Janardhan jointly released updated performance expectation guidance for engineers across the company on Tuesday. The memo explicitly stated that Meta "will not use AI adoption dashboards or token usage to assess employee impact."

Evaluation criteria rewritten: From token trading to actual achievements

The new guidelines have made substantial revisions to the Meta engineer career development framework.

The memo removed previous references to specific AI tools and instructions for using AI for specific tasks, instead focusing on work quality and actual impact as the core considerations.

The memo stated: "Managers should focus on output quality, work speed, and the complexity and scope of the problems they are responsible for."

The memo also revealed that 93% of Meta's code changes are currently being assisted by AI agents , but then added that AI adoption "is not the goal in itself." "The goal is better results: higher quality delivery and faster iteration."

Token trading phenomenon: a side effect of policy

This adjustment comes as an unexpected consequence of Meta's previous aggressive AI promotion initiatives.

Earlier this year, Meta began mandating that engineers use AI tools in their workflows and incorporated AI Token usage and AI-assisted programming into hundreds of metrics tracked in its performance management system.

Last year, Meta also tried to gamify AI adoption, tracking employee usage and encouraging increased usage.

This mechanism has given rise to unexpected behaviors: some employees equate high token usage with good performance ratings, resulting in the so-called " tokenmaxxing " phenomenon—the deliberate consumption of AI tokens to demonstrate their high level of use of AI tools.

The CTO had already issued a warning, and this memo is the official closing.

In fact, Meta's management did not become aware of this problem until today.

According to The Information, Meta's Chief Technology Officer, Andrew Bosworth, warned in an internal memo back in April that "no one should use AI tools just for the sake of using them," and emphasized that "token usage is not a measure of influence in any sense."

The new memorandum jointly issued by the two senior executives is a further institutionalization of the above position—completely shifting the assessment focus from "usage behavior" back to "work results".

A Meta spokesperson declined to comment.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.