Many MindedAmazon tells engineers to use AI, but not to chase the numbers
Amazon has published an internal memo on AI use that it sent to software engineers and their managers in September. It was written by Terk Terkowitz, a VP and Distinguished Engineer on Amazon's independent contributor technical track. The details come from an opinion piece by columnist Corey Quinn in The Register, dated 31 October 2026.
The memo follows an awkward episode. In May, Amazon shut down KiroRank, an employee-built leaderboard that ranked staff by AI token usage. Employees had created junk tasks to climb it, a behavior now called 'tokenmaxxing'. The memo says this conflicts with Amazon's 'frugality' principle. It calls input metrics 'deeply flawed' but still advises using them carefully, and it warns that people who don't adopt AI are falling behind. It also says work that doesn't benefit from current AI tooling is 'a finding, not a failing', to be documented and raised with a manager. And it suggests checking whether teams in the 90th percentile of AI adoption, the heaviest token burners, show a matching rise in delivery.
Why it matters: this is a labor cost of the AI transition arriving before any support structure does. Engineers are told to adopt the tools, are measured in ambiguous ways, and see job cuts around them. Quinn notes AWS bills Bedrock and Kiro customers by the token, so token consumption also drives revenue externally.
Caveats: this is a columnist's opinion, not a news report. Amazon's reported 250 new job cuts this week are described as 'reportedly', and Quinn says he can't confirm they relate to the memo. His view that the mixed messages are deliberate cover for managers is his interpretation.
Source: The Register