To demonstrate to management that they are "frequently using" AI, some Amazon employees are utilizing the company's internal tool MeshClaw to automatically generate a large volume of unnecessary AI activities, aiming to boost so-called "token consumption."
"Competition" Gives Rise to Absurd Scenarios Recently, Amazon has begun a large-scale deployment of its internally developed MeshClaw tool, which can create AI agents to connect to work software and perform tasks such as code deployment and email classification on behalf of users. However, some employees are exploiting it to execute redundant, or even entirely meaningless, commands.
The underlying cause is significant KPI pressure. Previously, Amazon set targets for developers, requiring over 80% of employees to use AI tools weekly, and began tracking token consumption via an internal leaderboard at the start of the year. One employee stated bluntly: "The pressure to use these tools is immense. Some people use MeshClaw just to maximize token usage."
Security Concerns and Management Reflection Although Amazon has informed employees that this data will not be used for performance evaluations and has warned managers against using it to measure performance, some employees indicate that management is still covertly monitoring this data. Regarding this, some employees have termed the phenomenon "token maximization."
Simultaneously, employees have also expressed security concerns about the tool. When performing tasks, MeshClaw requires user operation permissions, which poses a serious risk of the AI agent making "mistakes" or even executing unintended actions.
Comments