Artificial Intelligence

AWS Releases a Free Local Emulator It Says Was Written Entirely by Autonomous AI Agents

AWS Emulator lets developers and coding agents test AWS API calls on their own machines; AWS says its team has not written or reviewed any of the code, which is published on GitHub under a CC0 public domain dedication.

DailyMech Staff

Published

Last updated / Fact-checked on

AWS Releases a Free Local Emulator It Says Was Written Entirely by Autonomous AI Agents
Akela999 / Pixabay — Pixabay Content License

Rows of dark electrical cabinets with indicator lights along a data-center corridor

Illustrative photo of electrical cabinets along a data-center corridor; it is not an AWS facility or any product named in this article. Small white equipment labels on the cabinets are partly legible at full resolution, including "34UAT4-1," "34UAT4-2" and "34UAT4-3" on the left and one ending "UAT3-1" on the right; their second lines and the paper tags on the panel doors are not legible. No company logos are visible.

AWS on Wednesday released AWS Emulator, a free local emulator of AWS APIs that developers, test suites and coding agents can run on their own machines or in continuous integration, with no AWS account or deployment required. AWS says it has "no additional service fee"; users pay only for AWS resources when they deploy to AWS.

AWS says the emulator was written entirely by autonomous AI agents. Matthew Goldberg, an AWS senior principal engineer, wrote: "The emulator was written entirely by autonomous agents. Our team hasn’t written or reviewed any of the code." The source is on GitHub under the CC0 1.0 public domain dedication. The authorship claim is AWS's; DailyMech did not inspect the code.

At a glance

Item Detail Source/status
What it is Local emulator of AWS APIs for development, testing and CI AWS
Cost No additional service fee AWS
Who wrote it Agents wrote all code; team wrote or reviewed none AWS claim, unverified
Coverage More than 400 API models, more than 200 services GeekWire
License CC0 1.0; vendor/ code Apache 2.0 AWS; GitHub README excerpt
Known limits No IAM policy evaluation; single account; fidelity varies AWS
Runs on Linux packages; Docker on macOS and Windows AWS

How AWS says the agents built it

Goldberg says the project began as a weekend experiment with Kiro, AWS's coding agent, inspired by Anthropic researchers' demonstration that agents could autonomously build a C compiler. His three-step prompt asked Kiro to write tests based on the AWS CLI, iterate until they passed against real AWS, then build an in-memory implementation that passed the same tests.

According to Goldberg, Kiro returned a passable first version of Amazon SQS after 15 minutes. Over the weekend it built a rough version of the rest of AWS, with obvious gaps, and his team began using it that Monday.

Goldberg says the AWS API Models serve as the specification and the real AWS cloud as the reference implementation. One team of agents writes fidelity tests, real-world scenarios verified against real AWS; a second replays them against the emulator and files issues for failures; a third assigns one agent to fix each issue.

A merge gate blocks code that does not compile, and agents check coding standards, security and fidelity regressions before a change is merged. Goldberg wrote: "We’ve seen thousands of issues fixed every day as hundreds of agents concurrently work on the codebase."

Abib's AWS News Blog post says the agents run on Amazon Bedrock AgentCore Runtime instances with AWS CodePipeline and AWS CodeCommit, and that an independent security agent reviews changes adversarially. Goldberg's account also makes clear that people designed the prompts, guardrails and release pipeline, even though AWS says they did not write or review the resulting code.

What the emulator does

AWS says the emulator is written in Rust and installs from RPM or Debian packages on Linux; on macOS and Windows it runs in Docker, though AWS does not publish a ready-made image.

For operations it supports, AWS says the emulator returns the same results as AWS for valid calls and the same errors for invalid ones. Unsupported operations return 501 Not Implemented. State stays in memory until the emulator stops, and each copy gets its own port, keys and state.

GeekWire reported that the emulator is built from more than 400 API models covering more than 200 AWS services. AWS publishes per-service fidelity scores.

Limits AWS and analysts point to

AWS says the emulator does not evaluate IAM policies the way AWS does, so a request AWS would deny might succeed locally. Each emulator acts as a single account, with no cross-account or cross-Region behavior, and fidelity varies by service and feature. AWS says not to use it to test whether an application is secure. It sends anonymous usage and error data by default, which can be turned off.

IDC analyst Devin Pratt told InfoWorld, according to GeekWire, that the emulator does not yet simulate throttling, outages or regional differences in service availability, and does not sandbox the code it runs. DailyMech did not read InfoWorld.

GeekWire noted that LocalStack, another AWS emulator, this year put its free version behind an account requirement and limited it to noncommercial use. In an AWS post, LocalStack co-founder and CTO Waldemar Hummer said: "LocalStack is excited to work with AWS to build on this vision of a unified emulator ecosystem."

What to watch

Fidelity scores across releases, the details AWS has promised for re:Invent and Goldberg's planned follow-up posts.

How we reported this

DailyMech read AWS's What's New post, Abib's AWS News Blog post, Goldberg's aboutamazon.com post, AWS's dev.to post and GeekWire's report in full. The GitHub README was seen only in a search excerpt, and InfoWorld was not read. DailyMech did not install or run the emulator, did not review its code and contacted no one. The agent-authorship claim is AWS's.

Published and fact-checked November 11, 2026.

Sources

DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.

In Artificial Intelligence

From other sections

Sections