Meta has reportedly told workers that AI usage will no longer determine their performance reviews, just as employees test its Hatch agent. The useful shift is from counting tokens to judging outcomes, although inviting an agent into personal inboxes and calendars creates rather more consequential things to measure.
Watch Desk analysis
What happened
WIRED reports that Meta changed its review guidance this week, removing references to AI usage and the “AI Native” label. Three employees who received the internal announcement said the replacement language focuses on impact, which may be achieved with AI or by other means.
Meta spokesperson Tracy Clayton told WIRED that the company has always assessed contributions and that labels such as “AI Native” were never used for performance evaluation. Separately, staff have been testing Hatch on corporate devices for several weeks ahead of an expected public release. Meta declined to comment on that testing.
What changes
- Token counts leave the scorecard
Engineers were reportedly told that adoption dashboards and token consumption would not be used to assess impact. - Outcomes return to centre stage
The revised wording lets useful work count whether it involved AI or merely the unfashionable business of getting things done. - Hatch testing continues
The agent can browse the web and operate applications, making its access and mistakes more consequential than an ordinary chatbot’s. - Personal accounts raise the stakes
Some staff reportedly hesitate to connect private services because of privacy concerns and the risk of unintended actions.
Why it matters
Usage targets are a crude proxy for productive work. Remove the incentive and staff can choose AI where it helps instead of generating tokens to please a dashboard, the digital equivalent of looking busy near the photocopier.
Hatch makes that distinction urgent. An agent that acts across services may save real time, but adoption depends on trust, careful permissions and reliable recovery when it gets something wrong. A better scorecard will not solve those problems, but it may let employees surface them honestly.
Our read
Meta is right to judge results rather than AI consumption. Anyone evaluating workplace agents should copy that principle: measure useful outcomes, require least-privilege access and make opting out professionally safe. Token volume is a cost line, not a personality type.
What to watch
- Whether Meta publicly launches Hatch, and which accounts and actions it can access.
- What permission controls, audit trails and undo mechanisms accompany the agent.
- Whether Meta’s revised review language changes employee behaviour in practice.
Discussion spark: Should employers ever measure staff AI adoption, or only the quality and efficiency of the work produced?
Sources and evidence
- Source update (3 September 2026, 01:32 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.