Meta Pushes Its New AI Agent on Employees—but Eases Off on Tokenmaxxing

Nearly a year ago, Meta had stated that workers would be graded on their AI-driven impact, but in practice, this meant assessing how frequently they utilized chatbots and agents to enhance their daily tasks. The level of usage determined whether employees were labeled as "AI Native," "AI First," or "AI Enabled," according to a lawsuit filed in July by around two dozen workers who claimed Meta violated US antidiscrimination laws when they were laid off in May.

Certain health and family leave recipients contended they were unable to accumulate usage, leading to unfair penalties; however, Meta has refuted these allegations in the ongoing case. This week, new performance review guidelines have been released, replacing references to evaluating employees based on AI usage and their "AI Native" status with less stringent language, which states that "these outcomes can be supported by AI or other means."

It reaffirms Meta's emphasis on grading employees on their impact, with less focus on the method of achievement. Some Meta employees tell WIRED that the modifications are subtle yet beneficial, allowing them to avoid pressure to use AI when it may not be appropriate. A Meta spokesperson, Tracy Clayton, confirms to WIRED that the recent updates aim to underscore what has always been the case—Meta evaluates employees based on their contributions.

He adds that labels like "AI Native" were never utilized for performance assessment. Engineers across the company were informed this week that Meta "will not use AI adoption dashboards or token counts to evaluate impact," The Information reported on Wednesday. For months, some Meta employees have observed their colleagues repeatedly utilizing AI tools—for what they perceived as unnecessary purposes—to inflate internal usage metrics, such as the number of tokens, or AI units, consumed.

The tokenmaxxing trend peaked when a Meta worker crafted an internal leaderboard to monitor how much AI employees employed, categorizing them as "Token Legends." The employee subsequently took down the leaderboard in April after it was leaked online. A few months later, Meta introduced measures to regulate employees' AI usage. More recently, Meta has been encouraging, though not mandating, employees to employ its most advanced AI experiment, an agentic tool called Hatch that can autonomously perform actions on a computer.

Comparable to the well-known OpenClaw sensation, Hatch can browse the web and operate other applications. Employees have had the opportunity to test Hatch on their corporate devices for several weeks ahead of its anticipated public release. Some employees have been reluctant to connect Hatch to their personal email, calendar, and other non-work digital accounts due to privacy concerns, as well as the potential for unintended consequences, as mentioned by two current Meta employees familiar with these concerns.

Previously, Meta operated a project tracking employees' keystrokes and other activity on company devices to gather data for AI training, which has since been suspended, contributing to many employees' loss of trust in Meta's AI tools. However, the absence of fear regarding performance evaluations due to AI tool usage should help improve morale and potentially restore trust within Meta's workforce.

Some employees have already embraced Hatch for personal tasks like booking appointments and organizing their lives outside of work, describing it as effective; however, others express concerns over the environmental and economic implications of using substantial tokens and computational resources to automate tasks. Meta has declined to comment on Hatch testing.

Although employees might no longer aim to maximize token usage, Hatch still consumes more resources than typical chatbots and software coding assistants. Employees believe management still has an interest in witnessing their proficient use of AI tools. If Hatch ultimately enhances productivity as intended, there is a fear that Meta could implement further job cuts.

Last week, Reuters reported that Meta was preparing for another round of significant layoffs following the 8,000 job reductions in May, but this plan was abandoned due to an increase in software bugs related to AI reliance and slower-than-expected progress on agent development. Meta CEO Mark Zuckerberg has stated that further mass layoffs are not anticipated this year.