Amazon employees game internal AI leaderboards to meet usage targets
amazon staff are automating unnecessary tasks with the meshclaw tool to inflate token usage and meet an 80% weekly developer adoption target.
financial times reports detail how amazon employees are engaging in 'tokenmaxxing' by running unnecessary code deployments or email triage through the internal meshclaw agent tool. the gaming of the system aims to inflate token consumption metrics following a corporate mandate that 80% of developers use ai weekly. although amazon claims leaderboard stats do not impact performance reviews, workers cite competitive pressure and manager scrutiny as drivers for the artificial activity. this mirrors similar behavior at meta, highlighting the difficulty of accurately measuring ai productivity versus simple token volume.