Amazon Web Services has released a tool it says was written entirely by artificial intelligence. The software — an emulator that lets developers test AWS services locally, without a connection to the cloud — was produced by thousands of autonomous agents using Amazon's own agentic development environment, Kiro, and published on Wednesday.
"The emulator was written completely by autonomous agents," wrote senior Amazon engineer Matthew Goldberg in a blog post on how it was built. "Our team did not write or review a single line of code." That was possible, he said, only because the agents had to respect certain safety rules and the team trusted the bots not to cross those boundaries. Three teams of agents were set to work: testers, bug hunters and a repair crew.
The result functions, but looks far from finished. By Thursday afternoon the project's GitHub issue tracker listed more than 70 problems, concentrated in the simulated DynamoDB database and the serverless Lambda service. Concerns go beyond missing features. Because the emulator does not check IAM permissions the way real AWS does, a request that AWS would reject can succeed locally. "That is unfortunate," a developer wrote on Hacker News, saying he had hoped the official AWS product would outperform third-party tools such as LocalStack.
Reaction among programmers was mixed. One Reddit tester posted a bug list considerably longer than the list of queries that work, but rejected the claim that the tool is worthless: "The fact that something like this comes out in the first public version is very promising — it can only get better." He called the result nothing less than "a miracle".
The episode lands in the middle of a broader argument about how much software AI agents should be trusted to write — and how much of it humans still need to read before it ships.

