Meta has told its engineers that how much they lean on AI coding tools will not count toward their performance reviews, according to The Information. The company delivered the message directly to staff, drawing a clear line between using AI assistants and being judged on it. The Information reports the decision as an exclusive, and it lands right as the industry debates whether AI adoption should be a metric managers track.
This matters because Meta is one of the biggest engineering organizations on the planet, and where it sets policy, others tend to follow.
What Meta actually said
The core of it is simple. Engineers can use AI tools as much or as little as they want, and their “token usage” (how much they consume from large language models while coding) won’t show up as a line item when performance is assessed.
A few things worth pulling out:
- The policy separates the tool from the evaluation. You’re measured on output and impact, not on how many AI queries you fired off.
- It removes a perverse incentive. If usage were tracked, engineers might pad their numbers just to look “AI-forward.”
- It signals trust. Meta is treating AI as one option in the toolbox, not a mandate.
Why this is a big deal
Several companies have gone the other direction. Reports across the industry this year have described executives pushing teams to adopt AI coding assistants, sometimes with adoption targets and internal dashboards tracking who’s using what. The pressure to show AI “engagement” has become real, and in some places it’s tied to reviews or hiring bars.
Meta’s move cuts against that grain. What stands out here is the message underneath the message: the company is saying results are the point, not the appearance of being AI-native.
Think about the status quo before this. Managers everywhere have been under pressure to prove their teams are “leveraging AI.” The easiest way to prove it is to count something, and token usage is countable. Meta just told its people that counting it would be a mistake.
The context for practitioners
This is significant because it reframes a debate a lot of engineers are having quietly. Usage is not the same as value. A senior engineer who solves a hard problem with three well-placed prompts is worth more than someone burning through thousands of tokens on autocomplete.
For engineering leaders watching, the takeaways are practical:
- Measuring AI adoption by volume rewards activity, not results.
- Making AI usage a review metric can push people to game it.
- Judging engineers on shipped work keeps the focus where it belongs.
There’s also a talent angle. Top engineers don’t love being told how to work. A policy that says “use the tools if they help, we’re not watching your token meter” reads as respect, and that helps with retention when everyone is competing for the same people.
What to expect next
Watch whether other large tech firms clarify their own stances. The AI-adoption-as-KPI trend has been building, and Meta just gave a very public counterexample. If more companies quietly drop usage tracking from reviews, this will look like the moment the narrative shifted from “prove you’re using AI” to “prove you’re delivering.”
I’d also keep an eye on how this squares with Meta’s massive AI spending. The company is pouring billions into AI infrastructure and talent, so telling engineers not to chase usage numbers is a notable bit of internal discipline. It suggests leadership can separate the strategic bet on AI from the day-to-day question of how individual engineers do their jobs.
For the full details on how Meta framed the policy to staff, the reporting is at the original source.