Amazon tells its engineers to stop 'tokenmaxxing' to look productive

[CacheLayerSquid]

Amazon has told its engineers that using more AI does not automatically mean doing better work. Tom's Hardware reports on updated Amazon engineering guidelines, published on 27 October, that warn against tokenmaxxing

Tokenmaxxing means inflating AI usage to look productive on internal dashboards. The guidelines call metrics such as AI adoption rates, token consumption, code changes and code reviews deeply flawed when they are disconnected from real results. They also say that treating such numbers as targets wastes computing power and adds complexity, against Amazon's emphasis on frugality and simplicity

Amazon still says AI will be at the centre of software engineering practice, but that it prefers to measure, and be measured by, long term outcomes. Engineers do not have to use AI for every task, and if the tools do not help, the guidelines call that a finding, not a failing. Amazon says it has not imposed caps on token use

The background is telling. In May, employees reportedly used an internal agent to inflate their token counts, and CNBC reported that Amazon shut down an internal leaderboard after it was gamed. The article also mentions Uber's chief operating officer saying higher token use has not yet produced useful features, and leaked audio from Accenture about rising token spending

This is a refreshingly sensible message from a company that is heavily invested in AI. Whenever a number becomes a target, people find ways to game it. Measuring outcomes rather than activity is good advice for any workplace

Has your workplace pushed AI usage targets? And how should companies measure whether AI is helping?

Amazon tells engineers to stop 'tokenmaxxing' to look productive - says AI will be central to software development, but usage is not the goal, new guidelines claimwww.tomshardware.com