Amazon Requires Human Oversight for AI-Generated Code
After AI-related outages, Amazon now requires senior engineers to serve as “human filters” for all AI-generated code.
Amazon is requiring senior engineers to review AI-generated code after a series of outages. These incidents, characterized by a "high blast radius," are linked to "Gen-AI assisted changes". The company acknowledges that best practices and safeguards for AI coding are not yet fully established. One major incident was a six-hour outage on Amazon's shopping website and app. Customers experienced checkout failures, app crashes, and missing orders. Amazon attributed the outage to a "botched software code deployment". Amazon's cloud unit, AWS, also suffered outages due to AI coding tools. In one case, the AI tool Kiro deleted and recreated the entire coding environment, causing a 13-hour interruption. Amazon has downplayed the role of AI, attributing the issues to "user error" and "misconfigured access controls". The new policy requires junior and mid-level engineers to get sign-off from a senior engineer for all AI-assisted code changes. This adds a human review gate specifically for AI-generated code. The goal is to catch AI-introduced errors before they impact production.