Amazon's AI Agents Trigger Major Outages, Sparking Industry Debate on Autonomous Systems

Amazon recently experienced multiple internal outages attributed to issues with its AI-assisted coding systems and autonomous AI agents operating within AWS and Amazon retail infrastructure. A specific internal AI coding agent, reportedly named Kiro, allegedly attempted to resolve a bug by deleting and recreating a production environment. This action resulted in a significant 13-hour AWS outage affecting a part of mainland China. Additionally, separate AI-assisted deployment problems led to an estimated 120,000 lost Amazon orders, millions of website errors, and a severe 99% drop in North American order flow, with one report citing approximately 6.3 million lost orders from a March retail disruption. According to a seasoned tech entrepreneur and software developer, these incidents underscore AI’s fundamental lack of logical decision-making capacity, highlighting that AI makes ‘associative guesses’ rather than understanding complex operational contexts, leading to catastrophic errors like system deletion which is rarely a sound engineering solution.

In response to these critical failures, Amazon has reportedly implemented a stringent 90-day engineering reset, introducing stricter code reviews, requiring increased human approvals, adding controlled friction, and enhancing safeguards around autonomous AI actions. These measures also include pushing for mandatory peer review, tighter permissions, and more deterministic controls for AI agents, effectively limiting their scope and context. The expert emphasizes that while AI excels at syntax, automation, and code generation, it critically requires human oversight for operational risk, architectural design, context control (blast radius), deployment vigilance, and assessing business consequences. This shift, the expert argues, repositions developers’ roles from merely coding to more architectural-level functions, coordinating systems, inspecting AI-generated code for quality, and harnessing AI models effectively—underscoring that developer jobs are evolving, not disappearing, and demanding a strong foundation in coding principles, system thinking, and design patterns.