Meta ends AI-usage performance scoring while testing agent Hatch
Meta is formally ending a program that had graded employees on their use of AI tools, three employees who received the internal announcement this week told WIRED. Almost a year earlier, Meta had said workers would be graded on their "AI-driven impact," which in practice meant how much they used chatbots and agents, and usage correlated with labels such as "AI Native," "AI First" and "AI Enabled." About two dozen employees cited that grading in a lawsuit filed in July, alleging Meta violated US antidiscrimination law when it laid them off in May; workers on health and family leave say they could not accumulate usage and were unfairly penalized as a result. Meta has denied the allegations in the ongoing case.
New performance-review guidance issued this week drops references to "usage of AI" and the "AI Native" designation, replacing them with looser wording that says outcomes "can be supported by AI or other means," refocusing reviews on impact rather than method. Meta spokesperson Tracy Clayton told WIRED the change simply emphasizes what has always been true, that Meta evaluates employees on their contributions, and said labels like "AI Native" were never formally used in evaluations. The Information reported on Wednesday that engineers across the company were told this week Meta "will not use AI adoption dashboards or token counts to evaluate impact."
The program being wound down had a visible peak: a Meta employee built an internal leaderboard tracking AI usage, labeling top users "Token Legend." After the dashboard leaked, its creator took it down in April, and a couple of months later Meta moved to ration employees' AI usage.
Meanwhile, Meta has been encouraging, though not requiring, employees to test its newest agentic AI tool, Hatch, which can browse the web and operate other applications autonomously and is compared internally to the viral tool OpenClaw. Employees have been testing Hatch on corporate devices for several weeks ahead of an expected public release Meta has not dated. Some have hesitated to connect it to personal email, calendar and other accounts, citing privacy worries and fear it could disrupt something important; that hesitation is compounded by a now-paused Meta project that had tracked employees' keystrokes and other device activity to collect AI training data, which eroded trust in the company's AI tools. Others have already adopted Hatch for personal tasks like booking appointments, describing it as effective while worrying about the environmental and economic cost of the tokens and computing it consumes. Employees say Hatch still consumes more tokens than earlier chatbots and coding assistants, even without any intent to maximize usage, and believe management still wants to see them prove they can use AI tools effectively. Meta declined to comment on the Hatch testing.
Some employees fear that if Hatch does boost productivity as intended, Meta could follow with further job cuts. Reuters reported last week that Meta had been preparing another major layoff round on top of the 8,000 jobs cut in May, but called off those plans amid a rise in software bugs linked to AI reliance and slower-than-expected progress on agent development. Meta CEO Mark Zuckerberg has said no further mass layoffs are expected this year.
Key facts
- Meta's new internal guidance, issued this week, drops "usage of AI" and the "AI Native" designation as review criteria, replacing them with wording that outcomes "can be supported by AI or other means."
- The change follows a July lawsuit from about two dozen employees alleging Meta's May layoffs, tied to AI-usage grading, violated US antidiscrimination law; Meta denies the allegations.
- Even as evaluation criteria loosen, employees testing Meta's new autonomous agent Hatch, compared internally to OpenClaw, say their token consumption keeps rising.
- The prior program's peak was an internal "Token Legend" usage leaderboard, taken down in April after it leaked; Meta began rationing employees' AI usage a couple of months later.
- Reuters reported last week that Meta prepared, then called off, another major layoff round beyond the 8,000 May cuts, citing AI-linked software bugs and slow agent progress; CEO Mark Zuckerberg says no further mass layoffs are expected this year.
Why it matters
Meta is walking back one of the most visible attempts by a major tech employer to make AI use itself a graded job requirement, tracked down to token counts and adoption dashboards. Rolling that back weeks after a discrimination lawsuit over the program, while simultaneously pushing a new autonomous agent onto the same workforce, marks a real shift in how far AI-adoption pressure inside a big tech company can go, and where a company decides to pull it back.
Who it affects
Meta's engineering and product staff, whose reviews had been shaped by "AI-driven impact" labels such as "AI Native," "AI First" and "AI Enabled"; the roughly two dozen employees suing over the May layoffs, including workers on health and family leave who say they could not accumulate AI usage and were penalized for it; and the employees now testing Hatch on corporate devices ahead of its public release.
How to use it
Hatch is not yet public. Meta is encouraging, not requiring, employees to test it on corporate devices, and it can browse the web and operate other applications on its own, similar in kind to the viral agent OpenClaw. Some employees have already put it to personal use, such as booking appointments and organizing their lives outside work, and describe it as effective. Others are holding off on connecting it to personal email, calendar and other accounts over privacy concerns.
How solid is it
The account rests on three employees who received the internal announcement, an on-record comment from Meta spokesperson Tracy Clayton, and The Information's separate report that engineers were told the same thing this week. It confirms the review-language change and the ongoing Hatch testing, but does not quantify how many employees are testing Hatch, how much more token use it drives compared with earlier tools, or whether the review-policy change was a direct response to the July lawsuit.
Risks and caveats
Meta denies the lawsuit's allegations, and its spokesperson says AI-usage labels were never formally used in evaluations, a framing at odds with employees' account of a year-long program that graded them on exactly that. Trust in Meta's AI tools is already strained by a now-paused project that tracked employee keystrokes to collect AI training data, which is part of why some employees hesitate to connect Hatch to personal accounts. Employees also note Hatch still consumes more tokens than the chatbots and coding assistants it replaces, and some fear that if it proves productive, Meta could follow with the layoffs it prepared and then called off last week.
“will not use AI adoption dashboards or token counts to evaluate impact”
— Meta, in guidance given to engineers this week, as reported by The Information