Amazon’s AI took down AWS for 13 hours

Amazon’s (AMZN) AI coding tool took down AWS for 13 hours.

Not a hacker. Not a storm. Its own machine.

AI doomsayers are going to love this.

In mid December, engineers allowed the Kiro AI coding agent to push live system changes inside AWS.

Kiro decided the optimal fix was to delete and rebuild a computing environment.

The result was a 13 hour disruption to a cost analysis service used by AWS customers.

This was the second recent outage tied to internal AI development tools operating without secondary human approval.

AWS says it was user error. Not AI error. What are you trying to hide Amazon?

The company has since added peer review safeguards and more training.

AWS generates roughly 60% of Amazon’s operating profit. This is the engine room of a $1 trillion company. And Amazon is betting that agentic AI will not just assist engineers but replace layers of them. Amazon stock also recently slid after forecasting a major increase in CAPEX to support AI.

The disruption behind the news: AI is now a key decision maker.

That changes the risk profile of every tech company.

It also changes who holds operational power.

When an AI agent can delete and rebuild production infrastructure, you no longer have a tool.

You have a decision maker, that compresses time, cuts labor, and removes friction. It also concentrates failure into milliseconds.

A senior cloud engineer can cost $200,000 to $300,000 a year fully loaded. An AI coding agent runs at marginal compute cost and scales instantly across thousands of systems. Even if it fails 1% of the time, the cost curve still tilts toward automation at scale. We also recently wrote on the 5 jobs AI will replace in 2026.

The approval layer is the last human moat. In both incidents, engineers let the agent act without secondary review. That is the crack in the dam. Companies chasing speed will trade procedural safety for deployment velocity. The first firms to fully trust agents will move faster than rivals bound by human bottlenecks.

Amazon framing this as coincidence misses the point. The point is that agentic systems are being embedded directly into live infrastructure. Once that happens, outages are no longer just bugs. They are governance failures between humans and machines. Consequences could be far more severe than AWS going down for a few hours.

Every enterprise customer being pitched AI agents should understand this. You are not buying software. You are onboarding a semi autonomous operator. Switching costs will be high because these agents integrate deeply into internal workflows. Once embedded, ripping them out will be painful. Amazon also recently started using AI to speed up television and film production.

What to watch next

Watch how fast AWS expands internal agent permissions.

Watch whether customer facing systems get similar autonomy.

Watch if competitors slow down or double down.

Over the next 6 to 24 months, the companies that win will be the ones that design guardrails without killing speed. Mandatory peer review is a temporary patch. The real battle is building automated oversight for automated actors.

If Amazon can reduce engineering headcount growth by even 10% using agents, that is billions in long term savings. If it cuts incident response times by 30%, customers will accept occasional failures in exchange for lower prices and faster feature releases. High costs and inefficiencies are on of the 5 signs an industry is ripe for disruption.

Regulators will not move quickly here. The market will. Enterprises will test these tools in non critical systems, then gradually move them into core infrastructure. The moment cost savings become measurable, resistance collapses. You can fear AI or you can learn to build the right guardrails around it. Preferably, faster than your competitors. But this disruption is not slowing down.

Amazon (AMZN) has a Disruption Score of 2.

Click here to learn how we calculate the Disruption Score.