THE CRUNCH
Amazon has published a treatise by Terk Terkowitz, a VP and distinguished engineer, on the developer's role in the age of AI. The document was originally an internal memo sent to software engineers and their managers in September, and appeared publicly at the end of October. It stresses customer outcomes and engineering judgement, documents work that AI tools are not currently improving, and argues that teams should look at the 90th percentile of AI adoption to check whether heavy token use actually raised delivery. Its central reassurance is that "if your work does not benefit from current AI tooling, that is a finding, not a failing."
The memo also calls out "tokenmaxxing", the practice of spinning up junk tasks to inflate AI usage metrics, as being at odds with Amazon's frugality principle. That follows a cautionary episode from May, when Amazon shut down KiroRank, an employee-built leaderboard that ranked people by token usage, after staff gamed it with pointless tasks.
Commentator Corey Quinn, writing in The Register, reads the document as well intentioned but sees a mixed message: it concedes that input metrics are "deeply flawed", then advises using them carefully anyway, and warns that non-adopters are falling behind. In his view, a manager could cite the "finding, not a failing" line for a strong engineer who skips AI, or cite "falling behind" when performance management suits, with both readings fitting the stated policy.
Quinn also notes the awkward backdrop: Amazon reportedly announced another 250 job cuts the same week, and while the memo treats tokenmaxxing as an internal problem, AWS bills Bedrock and Kiro customers by the token. He is careful to say he cannot confirm the events are related, only that engineers reading the memo assume they are.