Sooner or later it had to happen. He vibe codingthe programming code created by artificial intelligence, is not perfect, and Amazon has proven it firsthand, twice. Your AWS cloud has gone down twice for hours, due to two errors in the code created by AI.
Amazon assures that the fault is a human error on the part of its employees, the error could have still happened without using AI. But some of the workers who have spoken with Financial Timesconfidentially, insist that AI went too far.
«The outages were small, but completely avoidable. The company has set a goal of having 80% of its developers use AI-assisted coding at least once a week.»
This was the failure of AI with AWS
Many programmers fear they are being used as guinea pigs: they train the AI and correct its mistakes, and when it improves, they will be fired.
According to the media, the first incident occurred in mid-December. Kirothe AI agent developed by Amazon to program, while it was generating code, decided to delete everything and start from scratch. This caused the AWS cloud to partially go down for 13 hours, mainly affecting China.
The power that AI already has to manipulate critical elements of the Amazon cloud is surprising, despite being a first-generation AI. According to Amazon, In order for Kiro to make changes, he must have permission from a programmerwho had given it to him. They also discovered that this engineer had more permits than those corresponding to his position. That’s why Amazon calls it a human error.
Even so, this incident reveals two problems of vibe coding. The first, that AI is unpredictable, it does not think like a human. It never occurred to any programmer to delete everything, with the network up and running, and start from scratch. The second, although a human must give permission to the AI to make changes, over time they can assume that the AI is doing it right and give it permission as a routine, until in an oversight, the AI damages something.
In the second incident, a chatbot called Q Developerwhich assists engineers in programming, caused a failure that caused outages in different systems.
These Amazon AWS cloud outages caused by AI-generated codecome just weeks after Amazon laid off dozens of thousands of employees as the company restructures “to meet the challenges of AI.”