AWS releases AI-built emulator for local cloud testing

A local stand-in for the cloud

Amazon Web Services (AWS) has released a new tool designed to simulate its cloud environment directly on a developer's personal machine. This emulator allows users to test software against more than 200 different AWS services without needing an active internet connection. By running locally, the tool removes the need for developers to pay for real cloud resources during the testing phase.

The project aims to solve a specific technical hurdle created by the rise of AI coding agents. Tools like Amazon’s Kiro and Anthropic’s Claude Code can write software much faster than human teams can verify it against live cloud infrastructure. Because these agents often check their own work by running code repeatedly, using the real cloud can lead to significant delays and high costs.

AWS is offering this emulator for free under a CC0 public domain dedication. This means the code is available for anyone to use without restrictions, and the company has made the full commit history available on GitHub. The company plans to share more details about the project at its upcoming re:Invent conference.

Built by agents, for agents

In a notable shift in development philosophy, AWS engineers did not write or review a single line of code for this tool. Instead, the company used autonomous AI agents to build the emulator from scratch. The project was inspired by a demonstration where a team of Claude agents successfully built a C compiler.

The agents work using AWS’s public API models, which provide the instructions on how each service is supposed to behave. They operate in three distinct teams on the Amazon Bedrock AgentCore platform. One team creates "fidelity tests" based on real-world scenarios, such as creating and deleting a message queue, while a second team runs those tests against the emulator to find failures.

The third team is dedicated to fixing the issues identified by the second team, with one agent assigned to each specific problem. Every change must pass through an automated gate where other agents verify that the code meets specific security and coding standards. According to Matt Goldberg, a senior principal engineer at AWS, hundreds of agents are now fixing thousands of issues every day.

The human role in AI coding

The development of this tool highlights a changing role for human engineers in an era of AI-driven software creation. Rather than writing every line of code, engineers are increasingly tasked with defining the specifications and the rules for success. Swami Sivasubramanian, a vice president at AWS, noted that the engineer's job is now to build the "specs, tests, and merge gates" that decide if the AI's work is correct.

Goldberg described the current state of AI agents as being like a kindergartner who lacks the competence to hold a crayon properly. He suggested that while agents have the intent to do the work, they require clearly defined areas to fill in, which he humorously compared to providing them with "coloring books." The goal is to provide the AI with a structured environment where it can succeed.

This approach mirrors a recent move by Amazon to rewrite its engineering playbook for an era where AI writes the majority of the code. By giving agents a way to test their own work against a local emulator, the company can accelerate the development cycle. This allows the AI to iterate on complex tasks, like rebuilding replication engines, in a fraction of the time it would take a human to do the same manually.

Current limits and competition

While the emulator is a significant step forward, it does have several technical limitations that users should keep in mind. According to IDC analyst Devin Pratt, the tool does not yet simulate specific behaviors like service throttling or outages. It also does not account for regional differences in service availability, and it does not sandbox the code that it runs.

To maintain transparency, AWS has made the backlog of known gaps public in a GitHub folder titled "things-that-do-not-work." This allows developers to see exactly where the emulator falls short of the real AWS experience. Because the tool is open and free, it serves as a distinct alternative to other existing solutions like LocalStack.

LocalStack is a well-known emulator that recently moved its free version behind an account requirement and restricted it to noncommercial use. By releasing the emulator under a public domain dedication, AWS is positioning its tool as a highly accessible option for the developer community. The project began as a weekend experiment but has quickly evolved into a functional tool used by the company's own teams.