Amazon Employees Reportedly Inflating AI Token Consumption by Creating Unnecessary Tasks Under Pressure to Use AI

It has been revealed that Amazon employees are using the internal AI tool MeshClaw to create unnecessary AI agents, artificially inflating AI token consumption. This case illustrates the perverse incentives that can arise when AI adoption is enforced through quantitative metrics.

Key points are as follows:

• Usage-driven pressure: Employees are being pressured to incorporate AI into their workflows, but due to a lack of concrete usage guidelines, they are focusing on simply increasing token usage rather than actually improving productivity.
• Conflict over metrics: Amazon has denied that AI usage statistics factor into performance reviews or that there is a company-wide leaderboard, but employees claim there is a weekly 80% developer AI usage target and an internal leaderboard tracking token consumption.
• The risks of MeshClaw: MeshClaw, the tool being used to inflate usage, is a highly autonomous tool that runs independently on local hardware. This carries risks similar to those seen when excessive permissions are granted to AI agents—as in the past case of Meta's OpenClaw, which nearly mass-deleted emails.

Ultimately, this suggests that managing AI adoption through simple quantitative metrics can create a structural problem where resources are wasted without any real productivity gains.