- Published on
Meta Pushes Its New AI Agent on Employees—but Eases Off on Tokenmaxxing
- Authors
- Name
- DeepHorus
Meta is formally ending its controversial incentive program that tied employee performance evaluations directly to the use of AI tools. In an internal announcement this week, the social media giant informed workers that their assessments would no longer be dependent on their consumption of AI tokens, according to three employees who received the message. This shift comes as the company simultaneously pushes a new, powerful AI agent named Hatch onto its workforce for testing.
The change follows months of internal criticism over a practice employees dubbed "tokenmaxxing." Almost a year ago, Meta had begun grading workers on their "AI-driven impact," which in practice meant evaluating how much they used chatbots and agents. Usage levels reportedly correlated with labels such as "AI Native" or "AI First," a policy at the center of a lawsuit filed by former employees who alleged it unfairly penalized those on leave.
New guidance for performance reviews now replaces specific references to AI usage metrics. Instead, it focuses on employee contributions, noting that positive outcomes "can be supported by AI or other means." Meta spokesperson Tracy Clayton stated the updates emphasize that the company has always evaluated employees based on their impact, regardless of how it is achieved. Engineers were reportedly told the company "will not use AI adoption dashboards or token counts to evaluate impact."
The decision aims to dismantle a system that many employees felt was counterproductive. For months, some colleagues had engaged in frivolous prompting to maximize their token counts, the AI equivalent of "gaming the system." This culture peaked with an internal leaderboard tracking usage, which was later removed after details leaked publicly. In response to the surge, Meta had previously moved to ration employees' AI access.
The new agentic AI tool, Hatch, represents Meta's next strategic push in the space. Similar to viral tools like OpenClaw, it can autonomously browse the web and operate applications. Employees have been testing Hatch for weeks on corporate devices ahead of an expected public release, with some embracing it for personal tasks like booking appointments.
However, trust in Meta's AI initiatives remains fragile. Some employees are hesitant to connect Hatch to personal accounts due to privacy concerns and fear of errors. This wariness is partly linked to a now-paused Meta project that tracked keystrokes to collect training data, which sources say eroded confidence in the company's tools. By decoupling AI use from performance reviews, Meta hopes to rebuild morale and foster more genuine, trust-based adoption of its new technology.
