Product1 distinct publisher3 min readPublished
Removing the token count concedes that it measured willingness to perform adoption rather than work delivered. That concession lands while about two dozen laid-off employees litigate the labels the metric produced.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Any performance metric an employee can move without producing something a colleague would notice measures willingness rather than work. That is the failure mode Meta's own staff described to WIRED: colleagues prompting tools repeatedly, or frivolously in their view, to push internal usage measurements up [8]. Engineers were told this week that the company "will not use AI adoption dashboards or token counts to evaluate impact," The Information reported on Wednesday [3].
The sharpest evidence that the number tracked presence rather than output comes from the plaintiffs. Workers on health and family leaves say they could not accumulate usage and were unfairly penalized for it; Meta has denied the allegations in the ongoing case [6]. Whatever a court decides about intent, the mechanism is plain enough: a metric that counts activity per review period reads a closed laptop as underperformance. And for the year it ran, "AI-driven impact" meant in practice how far employees used chatbots and agents in their daily work, as Business Insider reported at the time [4].
Timing carries weight here. The usage framing went in almost a year ago and came out this week, so call it eleven or twelve months of live use [20]. The leaderboard came down in April, and rationing of employee AI usage followed a couple of months after that, which puts the supply cap around June, after the May layoffs the July suit contests [19]. The rescission arrives after the filing rather than before it.
What Meta now has, by accident, is a cleaner read than any dashboard gave it. Hatch is being encouraged and not required [11], which means voluntary use is measurable with no grade sitting behind it. Some employees called this week's changes subtle but welcome, freeing them from using AI in situations where it made no sense [17]. Others have already picked up Hatch for booking personal appointments and organizing life outside work, and describe it as effective [21], which says something real about the tool and nothing about engineering throughput. Consumption keeps climbing regardless, because the agent burns more tokens than the chatbots and coding assistants before it [14], a fact that leaves little room to read the rise as enthusiasm.
For whoever has to roll one of these mandates out on Monday, there are two axes worth drawing: whether the metric moves, and whether the work changes. Metric up with work unchanged is the theatre quadrant, and a usage count cannot see its way out, because from inside the count real adoption and theatre are the same number. Work changed with the metric flat holds the person who used the tool once, to good effect, plus everyone who was out on leave. Reward the first quadrant and you will manufacture more of it. The measures with any content in them are narrower and duller: whether a new user reaches a useful output sooner than they did before, and whether a team is still using the thing eight weeks after the pilot's sponsor stops asking. A leaderboard cannot capture either one. That gap is close to the whole reason the leaderboard existed in the first place.
Ranked by verification strength, evidence, and original report placement.
In an internal announcement this week, Meta told workers that their performance evaluations would no longer be dependent upon how much they used AI tools, according to three employees who received the message and spoke to WIRED.
New performance review guidance unveiled this week replaces references to criteria such as "usage of AI" and the "AI Native" designation with looser wording caveating that "these outcomes can be supported by AI or other means."
Engineers across Meta were told this week that the company "will not use AI adoption dashboards or token counts to evaluate impact," The Information reported on Wednesday.
Almost a year ago Meta said workers would be graded on their "AI-driven impact," which in practice meant evaluating the extent to which they used chatbots and agents to improve their day-to-day work, Business Insider reported at the time.
For months, some Meta employees say, colleagues prompted AI tools repeatedly, or what they viewed as frivolously, to maximize internal usage measurements such as the amount of tokens they consumed.
One Meta worker created an internal leaderboard tracking how much employees used AI, labeling them with titles such as "Token Legend," and took it down in April after details of the dashboard leaked.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Meta's Hatch agent tops out at $199.99 a month, with DoorDash and Etsy behind the meter2 distinct publishers
product
Meta blocked the search terms and left the videos up: nine takedowns from a list of 20-plus2 distinct publishers
leadership
Project OT put dates on the layoffs before it put numbers on the agents1 distinct publisher
product
Meta wants up to $199.99 a month for an agent, and it is selling a meter1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One newsroom, two listings
The policy change is the sturdiest thing here: three employees who received the announcement, two quoted fragments of the guidance itself, and The Information independently reporting the line about dashboards and token counts. Everything downstream of it thins out — the token surge has no numbers, the rationing has no date, and Meta will not discuss Hatch. Worth noting that the second listing in our coverage is the same WIRED text under a different headline, so it corroborates nothing.
Real inside Meta, invisible outside
Two things have genuinely shipped, both behind the badge reader: guidance that reached engineers across the company, and an agent on corporate laptops for several weeks. Hatch has no public release, no user count, and no company statement; the only usage signal is staff saying their token draw keeps rising, months after that same draw was rationed.
The rule went, the pressure stayed
"Scraps the AI usage test" is accurate about the wording and a little generous about the effect — and WIRED undercuts it in its own copy twice over. Employees call the change subtle, and they still believe management wants to see them wield the tools; token consumption is reportedly climbing as Hatch spreads. Meta, meanwhile, insists nothing substantive changed because the labels never graded anyone. Something narrower than an incentive program ended this week: its documentation.
A rewrite with a legal audience
About two dozen laid-off employees are in court arguing the AI labels sorted who was cut, which gives "labels such as AI Native were never used for performance evaluation" a second audience beyond the press. The employees on the other side of the story are anonymous, watching for the next round of cuts, and describing a tool Meta declines to discuss weeks before launch. Nobody quoted here is disinterested in how the timing reads.
Firm on the text, soft on the meaning
We can be fairly sure what Meta wrote this week and roughly when the old criterion arrived. What it does to behaviour is another matter: one newsroom, anonymous employees on every conduct detail, an unresolved conflict between the company's account of the labels and the plaintiffs', and a timeline that hinges on the phrase "a couple of months later."