Amazon puts humans back in the loop as its retail website crashes from "inaccurate advice" that an AI agent took from an old wiki
Four high-severity incidents in one week, including a six-hour checkout meltdown, traced to AI.
Amazon's aggressive push to integrate generative AI into its engineering workflows has backfired, causing a series of catastrophic outages on its retail website. According to internal documents obtained by the Financial Times, four high-severity incidents occurred in a single week, culminating in a six-hour meltdown last Thursday that locked shoppers out of checkout, account information, and product pricing. An initial post-mortem document prepared for a senior leadership meeting explicitly identified 'GenAI-assisted changes' as a root cause for a pattern of incidents dating back to Q3. However, this damning reference was deleted from the final version of the document before the meeting took place, highlighting internal tensions over the AI failures.
The core technical failure involved AI agents providing engineers with 'inaccurate advice' sourced from an outdated internal wiki, leading to faulty code deployments that crashed the site. The emergency meeting, led by the senior vice president of e-commerce infrastructure, was framed as a 'deep dive' into the systemic issues. In response to the cascading failures, Amazon is now reportedly scaling back its reliance on these autonomous AI tools for critical systems. The company is reinstating human oversight—a 'human-in-the-loop' review process—for code changes affecting its core retail platform, marking a significant retreat from its full-speed-ahead AI adoption mandate for engineers.
- Four high-severity website outages in one week, including a six-hour checkout meltdown, were directly linked to AI-generated code changes.
- Internal documents cited 'GenAI-assisted changes' as a factor, but this reference was deleted before a senior leadership review meeting.
- The faulty AI agents pulled 'inaccurate advice' from an old internal wiki, prompting Amazon to reinstate human oversight for critical deployments.
Why It Matters
This incident exposes the real-world risks of deploying autonomous AI agents in critical business infrastructure without sufficient guardrails.