Amazon has released a new local AWS emulator called "AWS Emulator" developed entirely by autonomous AI agents. It enables the development and testing of AWS applications by running AWS services in a local environment. Notably, the project for AWS Emulator does not involve any coding or code review by human engineers, demonstrating the ability of AI agents to autonomously generate code, execute tests, and resolve issues.

AWS Emulator: a local emulator built 100% by autonomous AI agents
https://www.aboutamazon.com/news/aws/how-autonomous-ai-agents-built-aws-emulator

GitHub - aws/aws-emulator: AWS Emulator is a local emulator of the AWS APIs. · GitHub
https://github.com/aws/aws-emulator

The development of AWS Emulator was inspired by a study conducted by Anthropic researchers in early 2026, which demonstrated that AI agents could autonomously build a C compiler. One of the motivations for the development was that AWS has an "AWS API model" as a specification and an actual AWS cloud as a "reference implementation," and that the environment was already in place for agents to autonomously build a solution that accurately implements the specification.

We first gave the following simple prompt to Kiro: ・Create tests based on the AWS Command Line Interface (CLI)
・Run the tests against the actual AWS and iterate until they pass
・Build an in-memory implementation that passes the tests

As a result of the prompt, the initial version of Amazon Simple Queue Service (Amazon SQS) was generated in just 15 minutes. Further refining the prompt, the full AWS local emulator was built over the weekend. Since the result was sufficiently compelling despite some shortcomings, testing of the software began the following week.

The development of AWS Emulator is carried out by AI agents divided into three teams:

・Test suite management team
Creates tests to measure how faithful the emulator is to the actual AWS cloud. The test content is generated from the publicly available AWS API model and is validated on actual AWS. The test suite is also updated when agents discover new capabilities that can be executed on AWS.

・Problem detection team
Identifies inconsistencies within the emulator. Agents are instructed to re-run fidelity tests, report issues for failed tests, and further infer the causes of failures to group those with potential common causes.

・Problem Correction Team
For each issue, check out the code suspected to be the cause, identify the root cause, and edit the code until the issue is resolved. Code changes must pass through gates that verify compilation, coding standards, security, and the absence of regression in fidelity. The code is only merged into the mainline if it passes all these checks.

Through the collaborative approach of the three teams, hundreds of agents work simultaneously on the codebase, fixing thousands of issues daily.


The source code of AWS Emulator is publicly available on GitHub under the CC0 license and is continuously updated. Developers and researchers can observe through GitHub how AI agents have built the emulator through thousands of commits. The resolved issues are managed in the "issues/things-that-now-work" directory, while unresolved issues are managed in the "issues/things-that-do-not-work" directory in the public repository, so there are interesting discoveries to be made upon further exploration.


AWS plans to continue sharing detailed information about the multi-agent workflow architecture, methods for measuring fidelity, and approaches to ensuring quality and security regarding AWS Emulator.