Meta dropped AI adoption metrics from performance reviews
The company discontinued adoption dashboards and token-consumption scoring after legal challenges — while rolling out its internal Hatch agent more widely.
Meta has discontinued the AI-adoption dashboards and token-consumption scoring it had been using in employee performance evaluations, following legal challenges, according to Wired. It is simultaneously rolling out its internal agent, Hatch, more broadly.
Scoring employees on how much AI they used was always going to measure how much AI they used. That it took legal challenge rather than a review of the results to stop it is the part worth examining.
What the metric actually rewarded
Token consumption is an input. Tying evaluation to it creates an incentive to consume tokens, which employees can satisfy without producing anything.
The outcome data from Meta's broader push towards an "AI-native" workforce suggests roughly that. Reporting this month described a 220 percent increase in code changes and a 36 percent rise in delivered features — alongside a 40 percent increase in incidents.
Read carefully, that is a productivity story with the cost attached. Features shipped rose by roughly a third; things breaking rose by two fifths. Code changes tripled, which mostly means the same work is arriving in more, smaller commits. Whether the trade is worth it depends on what the incidents cost, and Meta has not published that.
What it does establish is that the constraint on shipping software was never how fast code could be produced.
Why it drew legal challenge
The reporting attributes the reversal to legal challenges rather than to the results, and the exposure is not hard to identify.
A performance metric that determines pay and promotion is an employment decision. If it is derived substantially from automated telemetry, several jurisdictions have views: Illinois and New York City regulate automated employment decision tools, the EU AI Act classifies employment-related AI systems as high risk, and the Dutch data protection authority's €825 million fine against Uber last month turned on automated decisions taken without meaningful human intervention.
A dashboard ranking engineers by token consumption, feeding into review outcomes, is close enough to that description to be worth a company's while to stop.
Hatch is the part that continues
The more consequential fact is in the second half of the sentence: Meta is expanding Hatch, its internal agent, at the same time.
So the company has not retreated from AI in engineering. It has stopped measuring individual adoption while continuing to deploy the tooling. That is the correct sequence — the tool either improves outcomes or it does not, and if it does, usage does not need to be mandated.
It also removes the noise. When usage is scored, adoption figures become unreliable as evidence, because employees are optimising the metric. Removing the metric restores the ability to tell whether the tool is actually being adopted.
Where this leaves the wider question
The industry is running three different bets on AI and the workforce simultaneously. Caterpillar is spending $100 million to train 118,000 existing employees, on the view that experienced operators redesigning workflows is what makes deployment work. UBS will require AI proficiency from graduate candidates for 2027. Meta measured adoption and stopped.
Only one of those has produced published outcome numbers, and they are Meta's: more features, more incidents, and a metric abandoned under legal pressure.
Meta has not said what replaces the dashboards in evaluations, or whether employees scored under the old system will have those assessments revisited.
Runs the newsroom. Rename this profile in the studio to your own byline.
Related
Every weekday, the AI stories that moved money or shipped code.
No cross-posting, unsubscribe anytime. See all newsletters