Meta Drops AI Usage Quotas for Staff But Rolls Out a Powerful New AI Agent

Key points

• Meta told employees this week that performance reviews will no longer measure AI tool usage or "token counts," the units of computing work AI systems consume.
• The old system had labelled workers "AI Native," "AI First," or "AI Enabled" based on how heavily they used chatbots and coding assistants.
• About two dozen employees filed a lawsuit in July 2025 arguing the old system unfairly penalised staff on health or family leave who could not rack up AI usage.
• Meta is now testing an internal AI agent called Hatch, which can autonomously book appointments, browse websites and control other apps on a worker's device.
• Meta reportedly shelved a fresh round of layoffs after software bugs increased when teams leaned too hard on AI-generated code.

For roughly a year, working at Meta meant one strange new pressure: use AI tools as much as possible, or risk a lower performance score. That era is now over, according to Wired AI, which spoke to three employees who received the internal announcement this week.

The old rules tied annual reviews to "AI-driven impact," a phrase that translated in practice to tracking how many tokens, the small units of work an AI model processes when answering a question or writing code, each employee consumed. Workers who hit high numbers earned badges like "AI Native." Those who did not could be marked down.

The system went to strange places. One employee built a voluntary internal leaderboard called "Token Legend" to track colleagues' AI consumption. It leaked internally, embarrassed the company, and disappeared in April. Meta then began rationing staff AI usage. This week's guidance quietly buries that whole chapter, replacing the specific "AI usage" language with a note that good outcomes "can be supported by AI or other means."

Why did the old system cause problems?

The measurement pressure pushed some workers to use AI tools even when the tools were not helpful, just to hit numbers. Employees on parental or medical leave could not accumulate usage at all, so they fell behind on the metric through no fault of their own. A lawsuit filed in July by roughly two dozen ex-Meta workers argues this amounted to illegal discrimination when Meta cut those employees in May 2025. Meta denies the allegations; the case is ongoing.

Meta spokesperson Tracy Clayton told Wired AI that labels like "AI Native" were never formally part of performance scoring, and that the company has always evaluated employees on their contributions, not their methods.

What is Hatch, and should employees be nervous about it?

Hatch is an AI agent, a type of software that can take multi-step actions on its own rather than just answering questions. Point it at a task and it can open apps, browse websites, send messages and book appointments, all without a human clicking through each step. Meta has been letting employees test Hatch on their work laptops for several weeks ahead of a planned public release.

Some staff are reluctant to link Hatch to their personal email and calendar. That wariness traces back to a previous Meta project that logged employees' keystrokes on work devices to gather AI training data. Meta paused that project, but the trust damage lingers.

Despite the hesitation, some workers already use Hatch for personal admin tasks like scheduling appointments. They report it works well. Their concern now is a different one: Hatch consumes far more computing resources than a simple chatbot, and many wonder about the environmental cost of automating small everyday tasks at that scale.

The broader worry in the workforce is about jobs. Meta reportedly drafted plans for another major round of cuts on top of the 8,000 roles eliminated in May, but pulled back after AI-assisted coding produced a surge in software bugs and agent development moved slower than expected. CEO Mark Zuckerberg has said no mass layoffs are planned for the rest of this year.