Amazon has been aggressively pushing engineers to adopt AI tools. At least 80% of developers are expected to use AI in their coding tasks at least once a week. However, recent events suggest that such a rapid rollout may have come at a cost.
As the Financial Times reported, Amazon Web Services was shut down for 13 hours in December after engineers allowed the Kiro AI coding tool to update code without supervision. Kiro decided that “deleting and recreating the environment” was the best solution. I think that’s one way to solve the problem.
It wasn’t a one-time thing. A follow-up FT report found that Amazon’s e-commerce business has been dealing with ‘thinking trends’ since the third quarter of 2025, sparking an in-depth company-wide meeting led by SVP Dave Treadwell.
Some employees were already skeptical about how useful these AI tools actually were in their day-to-day work, and these incidents didn’t exactly help build confidence.
How bad has it gotten?
Business Insider has obtained internal documents that provide a clearer picture of what actually happened. On March 2, 2026, Amazon’s AI coding tools contributed to an incident that resulted in 120,000 lost orders and 1.6 million website errors.
Three days later, on March 5, 2026, a separate outage led to a 99% drop in orders across North American markets, resulting in a loss of 6.3 million orders. These are numbers that will definitely show up on the bottom line of financial sheets, even for a company as big as Amazon.
What is Amazon doing to make sure this doesn’t happen again?
Amazon is now rolling out a 90-day safe reset for approximately 335 critical systems. Engineers should have two people review changes before deployment, use formal documentation and approval processes, and follow more rigorous automated checks.
The company claims this is a user error, not an AI error, and the same mistake can happen in any developer tool. That’s a fair point, but it doesn’t change the outcome.
Giving broad powers to artificial intelligence tools without proper oversight creates problems, and the scale of AI-generated code only amplifies the damage.