AWS releases AI-built emulator to simulate cloud services locally

A weekend experiment turns into a tool

Amazon Web Services (AWS) recently released a new tool that allows developers to simulate its cloud environment on a local machine. This emulator provides a way to test software without needing an active internet connection or paying for actual cloud usage. As geekwire.com reported, the project began as a simple weekend experiment by an engineer who prompted an AI coding agent to build a version of Amazon SQS.

The engineer used a tool called Kiro to generate a functional version of the message queuing service in just 15 minutes. By the end of that weekend, the agent had produced a rough version of the rest of the AWS suite. The team began using this prototype to test their software on the following Monday, leading to the official release of the emulator.

AWS is highlighting the fact that its engineers did not write or review a single line of code for this tool. Instead, they used AI agents to build the software from the ground up. This approach serves as a proof of concept for how AI can be used to maintain and expand complex technical infrastructure with minimal human intervention.

How the AI agents build and fix code

The emulator functions by using more than 400 API models that describe how various AWS services are supposed to behave. These models serve as the blueprint for the AI agents to follow. The agents work in three distinct teams running on Amazon Bedrock AgentCore to ensure the emulator stays accurate and functional.

One team of agents creates fidelity tests, which are real-world scenarios like creating a queue or sending a message. A second team runs those same tests against the emulator to find any discrepancies or failures. A third team consists of individual agents assigned to fix every specific issue found by the second team. Every change then passes through an automated gate where other agents check for security and coding standards.

This automated loop allows hundreds of agents to fix thousands of issues every day. The agents are constantly comparing their work against the real AWS cloud to ensure the simulation remains as close to the truth as possible. This process allows the tool to scale much faster than a human team could manually update every service simulation.

Solving the speed of AI coding

Modern AI coding agents, such as Anthropic’s Claude Code, have made it easier for developers to write software rapidly. However, this speed can create a bottleneck because teams often cannot test the code against the real cloud as fast as the agents can generate it. Testing in the cloud adds significant delays and costs because the agents must run the code and read the results repeatedly.

The emulator addresses this by letting agents run those loops on a developer’s own machine. By simulating the environment locally, the agents can check their own work instantly without incurring fees or waiting for network responses. This allows the development cycle to keep pace with the speed of AI generation.

AWS leadership suggests this represents a shift in the engineer's role. Instead of writing every line of code, engineers are moving toward building the specifications, tests, and merge gates. The goal is to provide the AI with a clearly defined "coloring book" of rules to follow while the agents do the heavy lifting of construction.

Current limits and open access

While the emulator is a significant step forward, it does have specific limitations. It does not yet simulate complex behaviors like throttling, outages, or regional differences in service availability. Additionally, it does not sandbox the code it runs, meaning it does not provide a restricted environment for execution.

AWS has made the project transparent by making the backlog of known gaps public in a GitHub folder titled "things-that-do-not-work." This allows the community to see exactly where the simulation falls short of the real cloud experience. The company released the tool under a CC0 public domain dedication, making it free for everyone to use.

The release comes as a notable alternative to other tools like LocalStack, which recently moved its free version behind an account requirement. AWS plans to share more details about the emulator and its capabilities at its upcoming re:Invent conference. For now, it stands as a public resource for developers looking to test AWS services locally without the overhead of the live cloud.