AWS releases emulator built entirely by autonomous AI agents

AI-driven cloud simulation

Amazon Web Services has launched a new open-source tool called AWS Emulator. This free software allows developers to simulate various AWS services directly on a laptop, a development server, or within a continuous integration (CI) pipeline. By running these simulations locally, engineers can test how their applications interact with the cloud without needing to create active AWS accounts or pay for resources.

The most notable aspect of this release is that the tool was built entirely by autonomous AI agents. AWS engineers did not write or review any of the code. The company utilized a system of guardrails to ensure the agents worked independently while staying within specific safety and quality boundaries.

The emulator is intended for prototyping and testing rather than production use. Because it stores all emulated resources in memory, they disappear as soon as the emulator stops. It is designed to help developers verify service interactions before they move to a live environment.

How the agent teams work

AWS organized the development process into three distinct teams of AI agents, each with a specific role to ensure accuracy. The first team creates tests based on publicly available AWS API specifications. These agents verify the tests against real AWS services to ensure they accurately reflect how the actual cloud platform behaves.

The second team takes those verified tests and runs them against the emulator to find discrepancies. When the emulator behaves differently than the real AWS service, this team identifies the failures and groups them by underlying cause. This systematic approach helps the agents pinpoint exactly where the simulation is falling short of reality.

The third team is responsible for fixing the bugs reported by the second team. Each agent in this group handles one specific problem at a time. Before any fix is merged into the main codebase, it must pass several automated checks for coding standards, security, and fidelity. If a change fails these checks, it is sent back to the agents for revision.

Developer workflows and access

Developers have four different ways to integrate the AWS Emulator into their existing workflows. They can install it as a Linux package, use it via Docker for cross-platform support on Windows and macOS, or run it in an embedded mode inside a Rust test process. There is also a dedicated path for using it through an AI coding agent.

One of the primary benefits for teams is the ability to run integration tests in a CI pipeline without needing an AWS account. This allows for automated testing of code that calls multiple services across the cloud environment. It also provides a way for developers to prototype applications locally using AI coding agents before they ever commit to a deployment.

The tool provides several limitations that developers should keep in mind. It does not evaluate AWS Identity and Access Management (IAM) or resource-based policies, meaning it cannot be used for security testing. Additionally, it does not keep resources separate by region or account, so users are advised to use only one region and one account for all requests within a single instance.

Fidelity and transparency

The accuracy of the emulator varies depending on the specific service and feature being tested. AWS checks the behavior of the emulator against the US West (Oregon) region, while global services are checked against US East (N. Virginia). To maintain transparency, AWS publishes per-service fidelity scores with every new release of the tool.

The source code and the fidelity tests are publicly available on GitHub. This repository includes every issue filed by the AI agents, organized into folders for both fixed and unresolved problems. This level of openness allows the community to see the ongoing work the agents are performing to improve the tool.

While the tool is free, it does send anonymous usage and error data to AWS by default. However, users have the option to disable this reporting in the settings. The emulator does not send any actual requests, responses, or specific resource data to AWS during its operation. The software is released under the CC0 public-domain dedication, making it freely available for public use.