Amazon warns engineers against treating AI use as a target

Amazon has cautioned software engineers against treating AI usage and coding activity as standalone measures of performance, saying such metrics can encourage unnecessary spending and complexity if turned into targets. Its revised engineering guidance instead emphasizes outcomes for customers, the business and teams while allowing the activity measures to inform decisions when used carefully. The guidance, written by Terk Terkowitz, a vice president and distinguished engineer at Worldwide Amazon Stores, was circulated internally in September and published publicly on October 27. It describes token consumption, adoption of AI tools, code changes and code reviews as flawed standalone indicators because they do not show whether the work produced valuable results. Amazon’s warning addresses “tokenmaxxing,” the practice of using AI tools heavily without regard to whether they help complete meaningful work. The company says the measures can still be useful alongside other evidence, but managers and engineers should apply care and judgment rather than pursue higher usage for its own sake. The guidance asks managers to examine teams with both low and high AI adoption. They should...