Amazon Web Services has released a free local copy of its cloud that no human wrote. AWS Emulator, launched on 11 November, recreates more than 200 AWS services on a developer’s own machine, built from more than 400 public API models, and runs offline with no AWS account and no service fee. Amazon says autonomous AI agents wrote and tested every line of the code, and the company has released it into the public domain under a CC0 dedication.
What AWS Emulator does
The emulator serves the same AWS programming interfaces on a local machine, so the AWS command line, software kits and coding agents connect to it exactly as they would to the real cloud. Developers can run code, test it and fix it without a network connection, without an AWS account and without paying for cloud time while they iterate.
That matters most for coding agents. An agent works in a loop, writing code, running it and reading the result before deciding what to change next. When the services it calls live in the cloud, every loop costs time and money. With the emulator on the same machine, the agent can check its own work as it writes.
How agents built it
The idea came from Matt Goldberg, a senior principal engineer on the AWS Developer Tools team. His hypothesis was simple: given a written description of how a service should behave and the real service to check against, an agent could build a version of that service on its own. For AWS, the written description is its set of API models.
Goldberg tested it with Kiro, Amazon’s coding agent, asking it to build a passable version of Amazon’s SQS message queue. It returned one in about 15 minutes. By the end of that weekend, Kiro had built a rough version of the rest of AWS, and Goldberg’s team started using it to test its own software that Monday.
Three teams of agents now build and maintain the emulator on Amazon Bedrock AgentCore. One team writes fidelity tests, end-to-end customer scenarios checked against real AWS. A second runs those tests against the emulator and files an issue for each failure. A third fixes them, one agent per issue, and hundreds of agents now fix thousands of issues a day.
Every change still passes through a merge gate, a deep bench of automated checks that includes several security agents doing adversarial reviews. The full commit history and every issue the agents have filed are public in the project’s GitHub repository, including a folder named ‘things-that-do-not-work’ that lists the behaviours that do not yet match real AWS.
Our Read
Read plainly, the tool is the smaller half of the story. A free emulator of 200 services is useful, and it undercuts LocalStack, whose free version moved behind an account requirement this year. But the real claim is the method: a production-scale codebase that no human wrote or reviewed, built by agents working from specs and tests alone.
That is also where the honest caveat sits. ‘No human reviewed the code’ is a selling point only if the guardrails hold. AWS is candid that they do not always: the emulator skips access-policy checks and cannot yet simulate throttling, outages or regional differences. Users should read the public gaps list before trusting it with anything that matters.
The deeper shift is in the engineer’s job. Swami Sivasubramanian, AWS’s vice president for agentic AI, says the work is now to write the specs, tests and merge gates that decide when agent work is correct. He learned it rebuilding a replication engine: four months by hand, about two hours with Kiro once it could test its work.
Why It Matters
If the method holds, it changes what a small team can ship. Building test environments for 200 services by hand would take years; agents did it in weeks, and the same loop now maintains the result. Every cloud vendor faces the same testing bottleneck for agent-written code, so the first one to solve it in the open sets the pace.
The immediate win is practical: a free, offline test bed for the cloud’s biggest platform, with no account and no bill. The longer question is trust. AWS is asking the industry to accept code no human has read, guarded only by tests and merge gates. The emulator’s public record of what works, and what does not, will decide that case.
Background
AWS Emulator arrives as Amazon rewrites its engineering playbook for an era in which AI writes the code. The project began as a weekend experiment inspired by an earlier Anthropic demonstration, in which a team of Claude agents built a C compiler. AWS says it will share more about the work at its re:Invent conference.
The emulator is also a bet on openness. Its code is public-domain, its commit history is public, and its known gaps are listed for anyone to read. That openness is what lets outsiders check the claim that agents can build reliable software, rather than taking Amazon’s word for it.
Source: AWS’s announcement and GeekWire’s report.
0 Comments