AWS Says AI Agents Wrote Its Free New Cloud Emulator and No Human Reviewed the Code

AWS just handed developers a free copy of its cloud to run on their laptops. The bigger story is who built it. According to the engineer behind the project, no human on his team wrote a line of it, and nobody read the code before it shipped.

AWS Emulator, released Wednesday, simulates AWS services on a developer's own machine so that people and AI coding agents can test applications offline without paying for the real cloud. GeekWire reports that it is built from more than 400 API models covering more than 200 AWS services. AWS is pitching the authorship as a feature, not a footnote.

"The emulator was written entirely by autonomous agents. Our team hasn't written or reviewed any of the code." Matt Goldberg, AWS senior principal engineer

That line comes from Goldberg's launch post on About Amazon, and it is AWS describing its own process. No outside party has audited the repository to confirm that claim, so treat it as the company's account, not a verified fact.

A weekend hack that grew a merge gate

Goldberg says the project started as a weekend experiment inspired by Anthropic's demonstration earlier this year of Claude agents building a C compiler. He gave Amazon's Kiro coding agent a three step prompt, and Kiro came back after 15 minutes with what he called "a passable first version" of Amazon SQS, the company's message queuing service. By the end of the weekend, he wrote, Kiro had built a rough version of the rest of AWS with obvious gaps, and his team started testing against it that Monday.

The production version runs on Amazon Bedrock AgentCore with three teams of agents. One writes "fidelity tests," real scenarios such as creating a queue, sending a message and deleting it, and checks each one against the live AWS cloud. A second replays those tests against the emulator and files issues for failures. A third fixes them, one agent per issue. Every change then passes through an automated gate where other agents review coding standards, check security and look for regressions. Goldberg says hundreds of agents now fix thousands of issues a day.

He is candid that the weekend demo was not the product. "There's a natural temptation these days to see results like those and believe that AI can build anything. Maybe it will eventually," he wrote, adding that shipping something customers can trust "still requires serious engineering effort." His favorite failure: an earlier batch of agents decided to build their own Python interpreter for AWS Lambda.

So the humans did not vanish. They moved up a level. Swami Sivasubramanian, AWS vice president of agentic AI and emerging tech, framed that shift in a guest post on CTO Werner Vogels' blog.

"When agents write the code and check their own work, the engineer's job is to give them room to work on their own and to build the specs, tests, and merge gates that decide when that work is correct." Swami Sivasubramanian, AWS vice president of agentic AI and emerging tech

It is the same bet behind Anthropic's claim that Claude now writes over 90% of its own code: humans design the guardrails, machines do the typing. AWS just went a step further and says nobody read the output.

The emulator also has clear limits. IDC analyst Devin Pratt told InfoWorld, as cited by GeekWire, that it does not yet simulate throttling, outages or differences in which services and features are available by region, and that it does not sandbox the code it runs. Those are exactly the messy failures that break production. AWS is at least open about the rest. The agents' backlog of known mismatches sits in a public GitHub folder called "things-that-do-not-work."

The code is released under CC0, a Creative Commons public domain dedication, so anyone can take it. That lands awkwardly for LocalStack, the best known rival emulator, which according to GeekWire this year put its free version behind an account requirement and limited it to noncommercial use. AWS says it will share more at re:Invent. Until someone independent digs through that repo, the most interesting claim in this launch rests on AWS's word.