AWS Emulator Was Written Entirely by AI Agents
By @sharedot · · 6 pages
- AI Frontier
- Agentic AI
- Open Source
- Cloud
AWS released an open-source AWS Emulator built end to end by autonomous AI agents, with no engineer writing or reviewing any of the code.
A cloud simulator no human ever coded
Amazon Web Services has released AWS Emulator, a free tool that simulates AWS services on a developer's laptop, a development server, or in a continuous integration pipeline — and the company says autonomous AI agents wrote the tool entirely, with no code written or reviewed by its engineers.
Senior principal engineer Matthew Goldberg said the team could stay hands-off because the guardrails they built gave the agents space to work independently and AWS the trust that they would not "color outside these lines."
There is no additional service fee; users pay only for AWS resources when they deploy applications for real. The tool is available via an AI coding agent, a Linux package, Docker (Windows, macOS, Linux), or an embedded Rust mode.
Three agent teams, hundreds working at once
AWS describes three specialized agent teams.
The first generates tests from publicly available AWS API specifications and verifies them against real AWS services — tens of thousands of tests covering actions like creating a message queue, sending and receiving a message, and deleting the queue — which can then be replayed without contacting AWS again.
The second team runs those tests against the emulator, reporting behavioral differences and grouping failures that share a root cause.
The third team fixes reported issues, one problem per agent, and every change must pass build checks plus agent reviews for coding standards, security, and fidelity before merging. AWS says hundreds of agents work simultaneously, fixing thousands of issues each day.
What it can and cannot do
Developers can test cross-service calls locally, run integration tests in CI without an AWS account or resources, and prototype applications with AI coding agents before deploying.
But the limitations are sharp: AIBase-style caveats aside, Help Net Security reports the emulator is not intended for production, does not evaluate IAM or resource-based policies as AWS does (so it must not be used to test security), keeps all resources in memory so they vanish when it stops, requires a single region and account per instance, and has untested long-running or load behavior.
Fidelity varies by service, with AWS publishing per-service fidelity scores each release. Telemetry is anonymous by default but can be disabled, and requests or resource data are never sent to AWS.
Agents shipping real software is the headline
The surprising part is not the emulator itself but the demonstration: a major cloud vendor shipped production-grade developer tooling whose codebase its own engineers never wrote or reviewed, governed purely by agent-enforced checks.
AWS even publishes the full agent-generated issue backlog on GitHub, sorted into fixed and unresolved folders, under a CC0 public-domain dedication.