Amazon reviews Kiro outage fallout

This is the enterprise AI coding debate in its most important form: not whether agents can write code, but whether they can touch live systems without expanding operational risk. Amazon’s rebuttal doesn’t erase the story — it underlines that “user error” in an AI-assisted workflow still demands stronger guardrails.

• –Amazon says the confirmed disruption was limited to AWS Cost Explorer in one region and did not affect core compute, storage, database, or AI services
• –The company added mandatory peer review for production access, which is a meaningful signal that agent-driven changes need tighter human checkpoints
• –For developers, the lesson is not to avoid coding agents, but to constrain permissions, shrink blast radius, and keep auditability in the loop
• –Kiro’s spec-driven pitch is about adding structure to AI coding, and this controversy shows why structure matters even more in operations than in feature work