The Race for AI Tokens Infects Amazon: The 'Tokenmaxxing' Phenomenon Grows and Some Start to 'Cheat'

At Amazon, an internal practice known as tokenmaxxing is spreading, born from the increasing pressure to utilize the artificial intelligence tools introduced by the company. According to various testimonies, some employees are exploiting new proprietary software to generate non-essential automated activities with the aim of increasing the number of tokens processed by AI models, a metric that measures the volume of data processed.

The protagonist of this initiative is MeshClaw, a platform developed internally and rolled out on a large scale in recent weeks. The tool allows workers to create agents capable of interacting with corporate software and performing tasks autonomously, such as managing emails, analyzing notifications, monitoring code deployments, and integrating with applications like Slack.

At the core of this phenomenon lies a corporate policy aimed at increasing the use of artificial intelligence. Amazon has set a goal for over 80% of developers to use AI tools weekly, also introducing internal rankings based on token consumption. Although the company clarified that this data will not be used in individual evaluations, several employees feel that managers continue to monitor these metrics closely.

Amazon, AI, and the Impact of MeshClaw on Worker Habits

According to some sources, the monitoring of statistics has created distorted incentives. In a competitive environment, some workers use MeshClaw to artificially inflate their AI activity level, even when not strictly necessary. Similar dynamics have also emerged in other large tech companies, including Meta.

Amazon plans to invest about $200 billion in capital expenditures by 2026, with most resources allocated to artificial intelligence and data centers. MeshClaw, inspired by the OpenClaw project, was created with contributions from over thirty employees. An internal document describes the system as an assistant capable of consolidating what it learned overnight and preparing emails and tasks before the start of the day.

However, concerns remain. Several employees have expressed doubts about the safety of software authorized to act on behalf of the user, fearing errors or unwanted operations. Amazon emphasizes that MeshClaw helps thousands of workers automate repetitive tasks, exemplifying the integration of AI and human work.