Picture a job where your day is spent firing at targets, and management tells you your value is measured in rounds expended. Naturally, you shoot nonstop. Then the ammunition starts running low, and a memo arrives explaining that nobody actually expects you to empty the magazine. At almost the same moment, someone walks over and hands you the minigun from “Terminator 2: Judgement Day.”
According to Wired, that is more or less the state of AI inside Meta right now.
The scoreboard is getting switched off
On Wednesday, The Information reported that Meta had informed its engineers that “AI adoption dashboards” and “token counts” would no longer be considered in their evaluations.
Drawing on anonymous sources within the company, Wired reports that Meta’s newly published performance review guidance strips out “usage of AI” as an assessment criterion. Instead, the guidance reportedly frames the outcomes it wants as ones that “can be supported by AI or other means.” A number of Wired’s Meta sources said it is a relief to no longer feel obligated to reach for AI at every turn.
It is a genuine about-face — and the backstory explains why.
How tokenmaxxing became a competition in the first place
For most readers, the concept arrived in March by way of a Kevin Roose column in the New York Times, which described how certain AI-enthusiastic employers — Meta and OpenAI included — were prodding staff to run up their AI usage. Thus, “tokenmaxxing.” Roose floated OpenClaw, the then-popular token-chugging agentic AI platform, as one likely catalyst.
Per Roose, Meta was “rewarding workers who make heavy use of A.I. tools and chastening those who don’t.” His reporting also described leaderboards, effectively turning AI extravagance into a game with a scoreboard attached.
Then came June, when the Financial Times reported that Google was throttling Meta’s AI tokens — a squeeze that forced some rethinking of the whole tokenmaxxing push. The bullets, it turned out, were never in unlimited supply.
The minigun is called Hatch
Throughout all of this, Meta has been working to field a competitive AI tool of its own.
Two years ago, in 2024, Mark Zuckerberg declared, “This is going to be the year when a highly intelligent and personalized AI assistant reaches more than 1 billion people, and I expect Meta AI to be that leading AI assistant.” Whether Meta’s ChatGPT-style app ever got there is far from clear.
My own experience with the Meta AI web app amounts to a single visit, for a story. TechCrunch’s Amanda Siberling apparently did the same for a piece of her own and fared worse: Instagram alerted her friends that she was using it, and they mocked her for it.
Which may be why Meta appears to be betting on a productivity tool rather than taking another run at the ChatGPT market. The Information reported in May that the company was building Hatch, its own answer to OpenClaw — in other words, a consumer-friendly take on precisely the sort of token-hungry agentic platform that gave rise to tokenmaxxing in the first place.
Wired reports that, even as the tokenmaxxing incentives appear to be winding down, Meta “has been encouraging—but not requiring—employees to use its most advanced AI experiment.” The experiment in question is Hatch. There is still no public release, though staff have reportedly had access “for the past several weeks,” according to Wired. The tool is said to consume more AI tokens than conventional chatbots and AI coding assistants.
Some staff are apparently fans. Others among Wired’s sources cite an April news cycle — when it surfaced that Meta intended to train AI on employee keystrokes — as grounds for keeping their distance. That program was later paused by the company.
Agents are still the house religion
Whatever the review rubric now says, leaning hard on AI agents remains embedded in Meta’s workplace culture. In a memo circulated earlier this week, AI chief Alexandr Wang said the company’s migration from Google Chat to Slack was intended to make agent use easier still.
Wang described Slack as “the strongest platform available today for agents.” While conceding that “not all of us are building agents today,” he added that “everyone at the company will benefit from a robust and useful agent ecosystem.”
What employees are actually worried about
What worries staff, Wired reports, is that a smooth internal rollout of Hatch will be followed by job cuts.
Reuters reported last week that Meta had drawn up plans to eliminate as much as 60% of its workforce in certain areas. Those plans never went ahead, the report said, because the AI software meant to absorb the work of the departing employees was too buggy.
That is the detail worth dwelling on. The cuts weren’t abandoned out of principle. They were abandoned because the replacement wasn’t ready.
We contacted Meta with questions about the Reuters report and about internal testing of Hatch. No response was forthcoming.