# Amazon Warns Managers Against Setting AI Use as a Productivity Goal

By John Kim

Canonical URL: https://www.tokenpost.com/news/business/42404
Published: 2026-10-28T06:29:03.000Z
Updated: 2026-10-28T06:29:03.000Z
Section: Business

> The revised role guidance urges daily AI use and recommends comparing teams at the 10th and 90th percentiles of adoption.

Amazon is urging software engineers to build AI skills while warning managers that AI use and code metrics can distort productivity measures and raise costs when treated as goals.

The revised software development engineer guidance, published Tuesday, Oct. 27, identifies AI token use, adoption rates and counts of code changes and reviews as flawed measures when disconnected from outcomes for customers, the business and teams. Amazon warns that using those inputs as targets can increase costs and software complexity.

The guidance says the measures can still be useful when interpreted carefully. Amazon recommends reviewing teams at the 10th percentile of AI adoption to identify barriers, and teams at the 90th percentile to see whether heavier use is matched by greater delivery. The comparisons are meant to put adoption figures in context rather than make usage itself the goal.

Terk Terkowitz, Amazon’s vice president and distinguished engineer for Worldwide Amazon Stores, wrote that the company favors measuring sustained results. “We believe that it is best to measure, and be measured by, long-term outcomes,” he wrote.

Amazon also urges engineers who have not adopted AI to learn the tools and use them in daily work. The guidance treats work that does not benefit from current AI tools as useful information about where those tools do not help. “If your work does not benefit from current AI tooling, that is a finding, not a failing,” Terkowitz wrote.

The updated guidance describes the software development engineer role as changing alongside AI. It calls for engineers to combine AI tools with human judgment and knowledge of the business, while evaluating work by its results rather than by raw usage or code activity.