AWS releases emulator written entirely by AI agents
AWS has open-sourced AWS Emulator, a local emulator of its cloud services whose code was written and checked by autonomous agents, with no human writing or review.
AWS has released AWS Emulator, a local emulator of AWS services that developers can run on their own machines to build and test AWS applications. In a post published on About Amazon, the team says autonomous AI agents wrote all of the code and that "our team hasn't written or reviewed any of the code." The source is on GitHub under a CC0 license and is updated continuously.
The project started as a weekend experiment. The author cites Anthropic's demonstration that agents could build a C compiler on their own. They then reasoned that AWS already had a spec, its public API Models, and a reference implementation, the live AWS cloud. Kiro, AWS's coding agent, produced a basic Amazon SQS emulator in about 15 minutes and a rough version of the rest of AWS over a weekend. The team started using it the following Monday.
The production system splits the work among three groups of agents:
- Test writers create fidelity tests from the API Models and check each one against real AWS before submitting it. The suite now holds tens of thousands of tests.
- Issue finders run those tests against the emulator, group failures that may share a root cause, and file issues. The emulator began as an empty server that accepted requests for all 438 API Models but did nothing.
- Fixers each take one issue. Their changes must pass a merge gate that checks the code compiles, follows coding standards, is secure and does not reduce fidelity.
AWS says hundreds of agents work at the same time and fix thousands of issues a day. The post does not say how much of AWS the emulator now covers or what it costs to run.