The e-commerce group had published team-wide statistics about its employees’ use of AI, but recently restricted access so that only employees themselves and managers could view their statistics. Managers are reluctant to use tokens to measure performance, according to a person familiar with the matter.
Meta workers similarly engage in so-called “tokenmaxxing” to improve their position on internal leaderboards.
The MeshClaw tool that some workers use to boost their stats is inspired by OpenClaw, which became a viral sensation in February. OpenClaw allows users to run agents locally on their hardware, including PCs and laptops.
According to people familiar with the matter, Amazon’s MeshClaw can embed code, launch emails and interact with apps like Slack.
The company said the tool “enables thousands of Amazonians to automate repetitive tasks every day” and is an example of “empowering teams” to test and adopt AI tools.
“We are committed to the safe, secure and responsible development and deployment of generative artificial intelligence for our customers,” he said.
According to internal documents, more than three Amazon employees worked on the internal tool. A recent note describing the bot said: “It daydreams overnight to consolidate what it’s learned, monitors your placements while you’re in meetings, and checks your email before you wake up.”
Many Amazon employees have said they are concerned about the security risks of an artificial intelligence tool authorized to act on a user’s behalf. This risks situations where the agent may make mistakes or perform unexpected actions.
“The default security posture appalls me,” said one employee. “I’m not going to let him go and just go about his business.”
© 2026 The Financial Times Ltd. All rights reserved. May not be redistributed, copied or modified in any way.
