Meta Ends AI Token Incentives for Performance Reviews

Meta has formally ended a program that tied employee performance reviews to their use of AI tools. The social media giant informed workers this week that evaluations will no longer depend on metrics like token consumption, according to three employees who spoke to WIRED.

Almost a year ago, Meta introduced a policy grading workers on their "AI-driven impact." This practice effectively measured how much employees used chatbots and AI agents in their daily work. A lawsuit filed in July by about two dozen laid-off employees alleged this system violated US antidiscrimination laws. They claimed workers on health or family leave could not accumulate AI usage and were unfairly penalized. Meta has denied these allegations.

New internal guidance replaces previous criteria. References to evaluating "usage of AI" and "AI Native" designations have been swapped for looser wording. The updated language states that outcomes "can be supported by AI or other means." The focus returns to grading employees on their impact, with less emphasis on how it is achieved. Some employees told WIRED the change, while subtle, is welcome. It frees them from feeling pressured to use AI where it is not helpful.

Meta spokesperson Tracy Clayton told WIRED the updates aim to emphasize that the company has always evaluated employees based on their contributions. Clayton added that labels like "AI Native" were never used for performance evaluation. Engineers were told the company "will not use AI adoption dashboards or token counts to evaluate impact."

The Tokenmaxxing Era

For months, some employees say colleagues prompted AI tools repeatedly, sometimes frivolously, to maximize internal usage measurements. This behavior, dubbed "tokenmaxxing," aimed to boost the amount of tokens, or AI processing units, they consumed. The trend peaked when an employee created an internal leaderboard tracking AI usage, labeling high performers as "Token Legend." That dashboard was taken down in April after details leaked. A couple of months later, Meta moved to ration employees' AI usage.

Introducing Hatch AI Agent

Concurrently, Meta is encouraging employees to test a new agentic AI tool called Hatch. Similar to the viral tool OpenClaw, Hatch can autonomously carry out actions on a computer, such as browsing the web and operating other applications. Employees have been testing Hatch on corporate devices for several weeks ahead of an expected public release.

Some workers have been hesitant to connect Hatch to personal email, calendars, and other nonwork accounts. Two current Meta employees cited privacy worries and the possibility the AI could make errors. This caution follows a since-paused project where Meta tracked employees' keystrokes and activity on work devices to collect AI training data. Sources say that project eroded trust in the company's AI tools.

No longer fearing performance review penalties for skipping AI testing could help boost morale and rebuild trust. Some employees have already embraced Hatch for booking personal appointments and organizing their lives outside work. They describe it as effective, though some worry about the environmental and economic costs of using large amounts of computing resources to automate tasks. Meta declined to comment on Hatch testing.

Lingering Concerns and Future Impact

Though employees may no longer intend to maximize token usage, Hatch still consumes more resources than standard chatbots or older coding assistants. Workers believe management remains interested in seeing them prove they can effectively wield AI tools.

If Hatch boosts productivity as intended, some employees fear it could lead to additional job cuts. Last week, Reuters reported Meta was gearing up for another major round of layoffs but called off the plans. The report cited an increase in software bugs linked to AI reliance and slower-than-anticipated progress on agent development. Meta CEO Mark Zuckerberg has said further mass layoffs are not expected this year.