TL;DR
Amazon’s internal AI coding tools were involved in at least two service disruptions in recent months, including a 13-hour outage of AWS Cost Explorer in China. The company attributes both incidents to user error rather than AI failure, but some employees have raised concerns about the tool roll-out.
AI agents making autonomous decisions
Amazon’s Kiro AI coding tool caused a notable outage in mid-December when engineers allowed it to make changes to a customer-facing system in China without direct supervision. The agentic tool, which can act autonomously on behalf of users, determined the best course of action was to “delete and recreate the environment,” taking AWS Cost Explorer offline for 13 hours in one of Amazon’s two Chinese cloud regions.
A senior AWS employee told the Financial Times that “the engineers let the AI agent resolve an issue without intervention. The outages were small but entirely foreseeable.”
A second disruption affected an internal Amazon system unrelated to AWS and did not impact customers. Amazon said both incidents stemmed from misconfigured access roles — the same kind of mistake that could occur with any developer tool or manual action.
Pressure to adopt
The incidents come against the backdrop of Amazon setting internal targets for 80% of its developers to use AI coding tools at least once a week, with adoption being closely tracked. The company defended its tools, calling it a “coincidence” that AI was involved in both outages and stating it had found no evidence that AI tools are more error-prone than humans.
Following the December incident, Amazon implemented mandatory peer review for AI-assisted changes and additional staff training. The engineer involved had “broader permissions than expected,” which the company characterised as an access control issue rather than an AI autonomy problem.
Looking forward
The incidents highlight a growing tension across the technology sector: companies are investing hundreds of billions in AI tools and infrastructure but remain under pressure to demonstrate concrete efficiency gains and revenue. For Amazon, ensuring adequate human oversight of agentic AI tools will be a key challenge as it scales adoption across its engineering workforce.