Meta has ended employee evaluations based on AI use and is testing the agentic tool Hatch

This week, Meta published new rules for employee performance evaluations, replacing earlier criteria such as “AI use” or the label “AI Native” with broader wording acknowledging that results can be achieved “using AI or other means”. According to The Information, engineers were told that Meta would not use AI adoption dashboards or token counts to assess contributions. Meta spokesperson Tracy Clayton told Wired that the changes merely emphasize the existing practice of evaluating employees based on their contributions and that labels such as “AI Native” had never been part of performance evaluations — but this claim by the company conflicts with earlier reporting and a lawsuit brought by former employees.

According to the lawsuit filed in July by approximately twenty employees laid off in May, Meta previously evaluated workers based on so-called “AI-driven impact”, meaning the extent of their use of chatbots and agents, and assigned them labels such as “AI Native”, “AI First” or “AI Enabled”; the plaintiffs claim that this violated US anti-discrimination laws because employees on medical or parental leave could not accumulate enough AI usage and were unfairly penalized for it. Meta denies the allegations in the ongoing case. According to employees, the practice led to a phenomenon known as tokenmaxxing, in which workers unnecessarily ran AI tools repeatedly just to increase their metrics; one employee even created an internal leaderboard with labels such as “Token Legend”, which was discontinued after details became public in April. A few months later, Meta introduced limits on employee AI usage.

At the same time, Meta is offering employees the experimental agentic tool Hatch for testing. It can independently perform actions on a computer, browse the web and control other applications — according to sources cited by Wired, it is similar to the viral tool OpenClaw. Employees have been testing it for several weeks on company devices ahead of an expected public release; using it is not mandatory at this stage. According to sources, some employees are wary of connecting Hatch to personal email, calendars or other private accounts because of the risk of data leaks and potential errors by the tool; distrust is also deepened by an earlier Meta project, since suspended, that tracked keystrokes and other activity on work devices to train AI. Nevertheless, some employees already use Hatch for personal tasks and describe it as effective, while also pointing to the environmental and economic costs associated with high token consumption and use of computing power.

Some employees fear that if Hatch does increase productivity, Meta could make further layoffs. Reuters reported last week that Meta had been preparing another wave of layoffs beyond the 8 000 positions eliminated in May, but postponed the plan because of an increase in software errors linked to reliance on AI and slower-than-expected progress in agent development. CEO Mark Zuckerberg said that no further mass layoffs are expected this year.