Amazon revises AI guidance for engineers, warns against token overuse
Amazon has revised its internal guidelines for software engineers and managers on using artificial intelligence, according to a memo posted on its website. The document, first circulated in September, discourages "tokenmaxxing", the practice of using AI as heavily as possible with little regard for output. It tells engineers to inform management when AI is not useful for their work. Terk Terkowitz, a vice president and distinguished engineer at WorldWide Amazon Stores, wrote that counting tokens and lines of code are flawed ways to measure productivity. An Amazon spokesperson said the company has not set caps on employee token usage.
Key facts
- Amazon shut down an internal leaderboard called Kirorank in May.
- The memo describes token usage, AI tool adoption rates and code changes as deeply flawed metrics.
- Employees had used AI on meaningless tasks to inflate token counts on the leaderboard.
- An unnamed company reportedly spent 500 million US dollars on Claude in one month.
Summarised with AI from 2 sourcesRead our AI policy
Why it matters
Shows a large technology employer moving away from raw AI usage counts as a measure of engineering productivity.
What changed
First published from CNBC and Fortune