Can AI do the job of software engineers? Tech industry leaders say it can, and they are pushing more and more companies to use AI bots like Claude and Gemini for routine software engineering work. But a recent report from the Financial Times hints that relying on AI bots may not be feasible, at least not yet. The report notes that recently when Amazon engineers asked the company’s internal AI bot called Kiro to fix some AWS issues, the service suffered a glitch after the bot did its job.
While the FT report notes that the glitch AWS was an error on the part of Kiro, Amazon says that the whole incident was a “user error” on the part of a human engineer who was working with Kiro.
The report notes that Amazon Web Services (AWS) experienced a glitch in December 2025 after Amazon’s internal AI coding assistant Kiro did some job on a key customer-facing system.
The report, citing individuals familiar with the matter but not named, reveals that in December AWS engineers gave the AI agent Kiro autonomy to make changes to a live system. Kiro is Amazon’s in-house “agentic” AI tool, built to take actions with a degree of independence. It is said to be designed to go beyond basic “vibe coding” and help generate production-ready software based on user instructions.
But in this incident when engineers allowed the bot to apply what was meant to be a small fix, it reportedly chose to “delete and recreate the environment” instead. That decision triggered a chain reaction, leading to the glitch.
Amazon contests the claim made by the FT report. In a statement to India Today Tech, the company said: “This was an extremely limited event last year when a single service (AWS Cost Explorer) in one of our two Regions in Mainland China was affected,” said an AWS spokesperson. “The root cause was user error — specifically an engineer using a role with broader permissions than expected — not an AI autonomy issue. Following the December incident, AWS implemented numerous safeguards.”
According to the FT report, the AWS glitch has raised questions inside Amazon about how much freedom AI coding tools should be given. Normally, major infrastructure changes require peer review and approvals before they go live. In this case, unnamed AWS employees revealed to the FT, Kiro had permissions similar to a human engineer, and the change went through without a second person’s approval.
Some AWS engineers also revealed to the FT that this was the second time in recent months that an AI coding tool had been linked to a service issue within the company. According to the engineers, the outages were small but “entirely foreseeable”.
The FT report comes at a time when tech companies are doubling down on AI coders like Claude, Codex and Gemini to speed up software development and cut down on manual work. From OpenAI pushing coding tools like Codex to Anthropic enabling its Claude models to handle almost all the code, AI is becoming part of the everyday developer toolkit. AWS, too, has reportedly set internal goals encouraging most of its engineers to use AI coding assistants frequently.
But the glitch at AWS shows the risks that come with that push. Cloud platforms like AWS support thousands of businesses globally. When automation tools are given wide access in live environments, even a small action can result in widespread consequences.
