Amazon Holds Mandatory Meeting After Vibe-Code Triggered Major Outages

A trend of GenAI-assisted (vibe-coded) changes triggered major outages at Amazon's ecommerce business, followed by a meeting on March 10, 2026, where engineers were summoned to attend by a top executive. According to leaked internal Amazon documents and employee testimonies, the meeting's agenda cited a trend of incidents and unsafe practices with a high blast radius and listed novel GenAI usage, for which best practices and safeguards are not yet fully established, as a "contributing" factor (can be traced back to code developed by generative AI).

The email from Dave Treadwell, Senior VP of ecommerce services, told employees: "Folks, as you likely know, the availability of the site and related infrastructure has not been good recently."

A 13-hour AWS outage in December 2025 happened because the Kiro AI coding tool was allowed to decide that the best course of action was to "delete and recreate the environment." Multiple Amazon employees confirmed to reporters that engineers had let AI agents resolve issues without intervention.

Following the December 2025 outages, Amazon implemented safeguards such as mandatory peer review for production access and staff training. As we can see, it had little effect.

Last week in early March, both Amazon's website and shopping app went down. For six hours, customers could neither check out and complete transactions, access account information, nor view product prices. Before that, there was a severe 15-hour AWS outage in October 2025 that forced multiple customers' apps and websites offline, including OpenAI's ChatGPT.

The Lack of Guardrails

On March 2, customers across Amazon marketplaces saw incorrect delivery times when they added items to their carts. It led to nearly 120,000 lost orders, according to Business Insider. Amazon's AI tool Q was one of the primary contributors that triggered the event. An internal document warned that because GenAI generates code so quickly, it is accidentally exposing vulnerabilities and proving that the company's current safety guardrails are completely inadequate.

This lack of guardrails was evident again on March 5. Another outage caused a 99% drop in orders across Amazon's North American marketplaces, resulting in 6.3 million lost orders. One key factor was a production change that was deployed without using a formal documentation and approval process called Modeled Change Management. There was no automated pre-deployment validation, so a single authorized operator could execute a high-blast-radius config change.

Now, Amazon is rolling out a 90-day, temporary safety guideline on top of their existing policies. The new policy targets approximately 335 "Tier-1" (money-making) systems or services that impact consumers. Amazon engineers must get two people to review their work before making any coding changes. Tier-1 system owners, as well as Director- and VP-level leaders, must audit all production code change activities.

"We will invest in more durable solutions including both deterministic [rules-based programming] and agentic [using AI to review the code] safeguards".
Dave Treadwell

Amazon Blames Human Error & Coincidence

Executives are mandating the use of AI tools to write code (setting targets for 80 percent usage) to increase speed and cut costs. The AI tools are known to sometimes make catastrophic errors, such as deleting the environment.

When the AI makes a mistake, executives (such as Amazon) officially blame "human error" or "misconfigured access controls" because the engineer was supposed to perfectly oversee, use a second person's approval before making changes and catch the AI's mistakes.

"This brief event was the result of user (AWS employee) error, specifically misconfigured access controls, not AI".
An Amazon spokesperson writes in a statement sent to The Register

"It was a coincidence that AI tools were involved, and the same issue could occur with any developer tool or manual action. In both instances, this was user error, not AI error".
Amazon writes in a statement to the Financial Times.

An Amazon spokesperson told Business Insider that only 1 incident reviewed on Tuesday was AI-related, and none of them involved AI-written code.

Internet users joke: Companies want all the productivity benefits of autonomous AI without taking any of the responsibility when that autonomy causes a disaster. Instead, they blame the human employee whom they forced to use the AI.

Amazon is spending $200 billion on data centers, cloud infrastructure, and AI development. To afford this, Amazon is laying off tens of thousands of workers, including 16,000 in January 2026 alone. Because there are fewer engineers left to do the same amount of work, Amazon's leadership is forcing the remaining employees to use AI coding tools to speed up their output. The remaining engineers are pushed to code faster using AI (vibe-coding).

"An I-shaped specialist, someone with knowledge in only one area, is becoming outdated. To solve hard problems, developers must look past their main skill".
Dr. Werner Vogels, CTO of Amazon

But AI code constantly requires thorough human review. Because Amazon fired so many people, there are not enough engineers with the time or capacity to double-check the AI's work. As the Financial Times reported, Amazon engineers complained that their teams had to deal with a higher number of "Sev2s" (major incidents requiring rapid response) specifically because of these recent job cuts.

Amazon tried to replace human engineers with AI to save money, but the AI is not good enough to work unsupervised.

Amazon has disputed the claim that headcount cuts were responsible for an increase in recent outages, according to the Financial Times.

The Danger of Vibe-Coding

Vibe-coding, in which you tell AI what you need and sit back, is on the rise. Anthropic released Claude Code in February 2025 and passed $1 billion in annualized recurring revenue by November. They also released Claude Cowork, a version of Claude Code built for everyday knowledge work instead of just programming. 

Both Microsoft and Google state that over a quarter of their code is now written with AI. Engineers at Anthropic and OpenAI say that nearly 100% of their code is AI-written.

The Amazon incident has opened up a conversation about whether vibe coding could lead to badly designed software. Third-party code reviewing software platforms report that, out of hundreds of pull requests (engineers speak for bug fixes or changes in code), AI code had 1.7 times more issues than human code.

"We can't just pull a lever on your IDE and hope that something good comes out. That's not software engineering but gambling. Remember, the work is yours, not the tool's. AI can generate code faster than you can understand it. That gap allows software to move toward production before anyone has truly validated what it actually does".
Dr. Werner Vogels (CTO of Amazon) at AWS re:Invent 2025

"LLMs sometimes make assumptions you do not realize they are making. If you do not tell it explicitly that something needs to work for multi-threading, for example, it may produce the minimum version that works, but when it is scaled or put into production, it could crash.
It is very hard to resist vibe coding nowadays. When your peers are using it and coding faster, it is hard to resist. If you cannot keep up with the speed, it becomes difficult to collaborate".
Anni Chen (Amazon tech lead)

"There is a capability reliability gap in AI-assisted coding. Systems can look impressive at the moment, but reliability only shows up over time, after varied real-world use".
Lilly Ryan (cybersecurity consultant and software historian)

There is a lot of software that we can automate, and that does not matter so much, and we do not need the best quality assurance for this. Then there is software where we want to be careful, and we want to keep the human in the loop."
Christian Kästner (associate professor of software engineering at Carnegie Mellon University)

AI Code Cleanup Company

Belitsoft offers a solution to the vibe-coding risks (lack of human oversight, security flaws, and scalability limits): agentic AI software development with senior engineer supervision.

Our Senior, Staff/Lead, and Principal Engineers, not AI, take full responsibility. Bringing AI coding tools into professional software development means that every AI-generated line still goes through human engineer review, automated tests, and security scans.

Our senior engineers determine exactly how the code should look and behave, and check whether any AI suggestions comply with those standards. After the AI adds or modifies code, our team runs tests and reviews to catch any new bugs or unexpected side effects immediately.

Our most experienced engineers review every piece of AI-generated code, look for hidden mistakes, and adjust the output so it integrates seamlessly with your existing system.