Amazon has a question for its engineering teams that are slow to adopt AI

FILE - The Amazon logo is pictured at the Amazon Robotic Sorting Fulfillment Center in Madison County, Miss., Aug. 11, 2022. (AP Photo/Rogelio V. Solis, File)

Amazon has asked its engineering leaders to take a closer look at teams that are slow to adopt AI. In a memo sent to software engineers across the company and published on its website on October 27, Amazon recommended "paying special attention to the 10th percentile" of AI adoption, finding teams that "aren't embracing AI effectively and understanding why.

"The note accompanies Amazon's updated role guideline for software development engineers (SDEs), drawn up by its Distinguished and Senior Principal Engineers. It also points leaders to the other end of the scale, the 90th percentile, where teams use AI heavily, to check whether their delivery has gone up to match. Shared by Terk Terkowitz, VP and Distinguished Engineer for Worldwide Amazon Stores, the memo says AI will sit at the centre of software engineering, even as the company admits it is still early days.For two decades, Amazon kept specific technologies out of its SDE guidelines, even as cloud, microservices and serverless reshaped the job. Its senior engineers debated whether AI was any different and concluded it is changing the work "more quickly and more fundamentally" than those earlier shifts.

Amazon admits its AI usage metrics for engineers are flawed

To gauge AI use, Amazon has been relying on input metrics such as token usage and adoption of AI tools, along with throughput numbers like code changes and reviews.

The company calls these "deeply flawed" in the memo, since they are disconnected from what actually matters.It also warns that treating them as targets could push up costs and complexity. One line stands out: "tokenmaxxing is at odds with frugality," a reference to one of Amazon's leadership principles. Measuring lines of code, it adds, clashes with another, invent and simplify.Still, Amazon says these numbers track real output well for now.

Teams that have embraced AI the most, which it calls frontier teams, are delivering more, while teams and individuals not adopting AI are falling behind.

Amazon tells engineers it is fine if AI does not help their work

The memo is not all push. Engineers whose work does not benefit from current AI tools are told "that is a finding, not a failing." They are asked to document why, share it with peers and take it to their managers. Amazon says an engineer who can explain where AI falls short is showing the judgement it wants.Managers get homework too. They are expected to use AI on their own production work, and be able to describe one task where it helped and one where it did not.The company also wants engineers closer to customers. It admits some engineers and managers drifted away from the business as Amazon grew, and says those nearest their products are getting the most out of agentic software development.