Superpower Daily
LangChain Adds Agent Payments With Spending Caps, but Settlemen...LlamaIndex and Kaggle Launch a Document Benchmark for Cost and...Mostik Links AI Models Through Their Weights, Claiming a 20x Co...Lyte Raises $165M at $1.6B Valuation for Robot PerceptionAbliteration Removes Refusal Mechanisms for Offensive Cyber WorkNew Mexico Lets Districts Seek Approval to Leave K-2 AI Reading...LangChain Adds Agent Payments With Spending Caps, but Settlemen...LlamaIndex and Kaggle Launch a Document Benchmark for Cost and...Mostik Links AI Models Through Their Weights, Claiming a 20x Co...Lyte Raises $165M at $1.6B Valuation for Robot PerceptionAbliteration Removes Refusal Mechanisms for Offensive Cyber WorkNew Mexico Lets Districts Seek Approval to Leave K-2 AI Reading...
Businesspublished

Meta Drops Token Counts From Reviews While Employees Test Hatch

Meta is separating performance assessment from AI activity just as it puts a more autonomous tool in front of employees. Whether workers trust Hatch with personal accounts remains a separate test.

By 2 min read
Meta Drops Token Counts From Reviews While Employees Test Hatch
Meta Drops Token Counts From Reviews While Employees Test Hatch
Sign in to save and react

Listen to this story

The audio brief

About 1:34
0:001:34
Read transcript
Meta is removing AI usage from employee performance reviews just as workers test Hatch, an agent that can browse the web and operate other applications. The timing creates a clear split: employees are judged on what they accomplish, not on how much they use AI, while Meta is asking them to decide whether a more autonomous tool belongs in their digital lives. Under updated review guidance, references to AI use and labels such as “AI Native” are gone. Engineers were told that token counts and AI-adoption dashboards will not measure impact. Meta spokesperson Tracy Clayton said the company has always evaluated people on their contributions, and that the AI Native labels were never part of performance evaluation. Hatch has been running on corporate devices for several weeks, ahead of a public release that has not yet been announced. Some employees are reluctant to connect personal email, calendars, and other accounts. The concern is practical as much as private: an agent that can take actions could also make a consequential mistake. Some workers have used Hatch for personal appointments and organizing their lives, but they say it consumes more tokens than standard chatbots and older coding assistants. The change also follows a damaged trust environment. Meta removed an internal token-use leaderboard in April, later rationed AI access, and previously paused keystroke and activity tracking after employees objected. The question now is whether removing usage pressure will make workers trust Hatch—or whether fears about privacy, errors, and job cuts will still outweigh its convenience.

Story brief

3 key points

Meta is separating employee evaluation from AI-consumption metrics just as workers trial Hatch, an agent that can browse the web and control other apps. Updated review guidance removes AI Native labels, token counts, and adoption dashboards as measures of impact, though the company says contribution—not AI usage—was always the standard. Hatch has been tested on corporate devices for several weeks, but privacy,...

  1. 01

    Hatch has been tested internally for several weeks, with a public release expected but not yet announced.

  2. 02

    Employees are hesitant to connect personal email, calendars, and other accounts because an agent could make consequential mistakes.

  3. 03

    Meta removed an internal token-use leaderboard in April and later rationed employee AI access.

Meta has told employees that performance reviews will no longer depend on how much they use AI tools. At the same time, staff are testing Hatch, an agent that can browse the web and operate other applications, creating a sharper divide between how work is judged and the tools workers choose to use.

The new guidance replaces references to AI use and labels such as AI Native with language saying outcomes can be supported by AI or other means. Engineers were also told that AI-adoption dashboards and token counts will not be used to evaluate impact, according to The Information. Meta spokesperson Tracy Clayton said the company evaluates employees on their contributions and that AI Native labels were never used for performance evaluation.

Hatch asks for a different kind of confidence

Employees have tested Hatch on corporate devices for several weeks ahead of an expected public release. Some are hesitant to connect it to personal email, calendars and other accounts because of privacy and reliability concerns, including the possibility of an important mistake.

Those concerns sit beside a recent trust problem. Meta previously paused a project that tracked keystrokes and other activity on work devices to collect AI-training data; employees said that project eroded trust in the company’s AI tools.

Less pressure, but not fewer stakes

The shift follows months when some employees said colleagues repeatedly prompted AI tools to maximize usage measures. An internal token-use leaderboard was removed in April, and Meta later rationed employee AI use.

What employees are weighing

  • Some employees have used Hatch to book personal appointments and organize their lives outside work.
  • Workers say Hatch consumes more tokens than standard chatbots and older coding assistants.
  • Some employees fear that productivity gains from Hatch could lead to additional job cuts.

The revised review system removes AI-use measurements from performance evaluation. It leaves a more practical question for the Hatch trial: whether workers see an agent with access to their digital life as useful enough to trust.

Sources

  1. wired.comMeta Pushes Its New AI Agent on Employees—but Eases Off on Tokenmaxxing

Related coverage

Continue reading