趋势解读："Tokenmaxxing" spreads at Amazon as employees game internal，评估 LLM Agent 表现

Amazon employees are automating unnecessary tasks just to climb internal AI leaderboards. The in-house tool "MeshClaw" lets employees create AI agents that can trigger code deployments, triage emails, or interact with apps like Slack. But according to the Financial Times, staff are deliberately using the software to artificially inflate their token consumption. "There is just so much pressure to use these tools," one Amazon employee said to the FT. "Some people are just using MeshClaw to maximise their token usage." The background: Amazon has set targets for more than 80 percent of developers to use AI each week, and earlier this year began tracking token consumption on internal leaderboards. Ad Officially, the numbers don't factor into performance reviews. But another employee disagrees: "Managers are looking at it. When they track usage it creates perverse incentives and some people are very competitive about it." Meta employees have engaged in similar " tokenmaxxing ." Ad DEC_D_Incontent-1 As a metric for actual productivity gains, token consumption is of little use. We took a closer look at the challenges of meaningfully measuring AI-driven productivity in the latest edition of our Frontier Radar . Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.