AWS 开源 AWS Emulator，可在本地模拟数百项云服务

Amazon Web Services Inc. is turning its public cloud computing platform on its head with the launch of a new service called AWS Emulator.

It’s a free, locally hosted software suite built by autonomous artificial intelligence agents that can replicate hundreds of the company’s cloud services on developer’s workstations. In a blog post, AWS Senior Principal Engineer Matt Goldberg said that the entire service was designed and written exclusively by AI agents, with not even a single line of its code written or reviewed by a human engineer.

AWS Emulator is designed to help customers overcome a major bottleneck in the application development process that has arisen with the adoption of AI coding agents such as AWS Kiro and Claude Code. Those agents are generating line after line of code faster than human programmers can blink, and the sheer velocity of that production is causing real issues for the teams tasked with verifying it before it can ship in production.

All AI-generated code needs to be validated against real cloud infrastructure, but doing that in the real AWS cloud is becoming all but impossible given latency and cost considerations. Though there are some existing third-party AWS emulators that developers can use, those services cover only a small fraction of AWS’ sprawling catalog of cloud services and quickly drift out of sync as those services evolve. With AWS Emulator, developers now have access to a free-to-use, locally hosted emulator that developers can use to run multiple iterative test loops on their own machines, Goldberg said.

Built by AI agents, for AI agents

The new tool was the brainchild of Goldberg, but he was actually inspired by the creation of AWS’ foundational Paxos-based distributed locking service, called “Alf,” which still underpins core cloud services such as Amazon S3 and DynamoDB.

In a second blog post, Swami Sivasubramanian, AWS’ vice president of agentic AI and emerging technologies, explained that when he and his colleagues first built Alf more than 20 years ago, they could test it only by simulating its entire environment inside a Java virtual machine built manually over several months.

Sivasubramanian explained that his team would have to inject faults and isolate concurrency bugs manually before they could deploy new services in production. But as AWS’ cloud has scaled up, it’s no longer feasible to hand-craft a simulation of its entire platform.

This is the conundrum that inspired Goldberg to see if that task could instead be done by AI agents. After seeing research that showed how AI agents were able to build a C code compiler from scratch, he concluded that they should also be able to replicate any AWS service if given a formal specification and reference implementation.

So he set about doing so, using AWS’ public Smithy API models as the specification and the live AWS platform as the ground truth. He began by prompting AWS Kiro to rebuild the Amazon Simple Queue Service, and it generated a functional replica of that service within just 15 minutes.

Seeing the potential, Goldberg and his team then set about organizing dozens of autonomous agents running on Amazon Bedrock AgentCore into three separate teams. The first is tasked with generating thousands of end-to-end fidelity test suites from API models and verifying them against the live AWS cloud. The second runs the tests against the emulator to try and find any bugs or discrepancies, and if any are identified, those agents will automatically file a GitHub issue.

The third team was tasked with creating and implementing the fixes for those bugs. As a final step before the code enters the GitHub repository, it goes through an automated merge gate that’s supervised by adversarial security agents to check for any vulnerabilities.

A new role for software engineers

AWS Emulator has been open-sourced under a Creative Commons CC0 public-domain dedication license, together with its raw commit history and a public backlog of known discrepancies filed under “things-that-do-not-work.”

Sivasubramanian said AWS Emulator illustrates the dramatic evolution of the role of the software engineer. With AI agents now responsible for the grunt work of writing, testing and debugging the actual code, humans are now becoming “systems orchestrators,” tasked with defining the boundaries, architecting the test specifications and designing the guardrails that guide the work of those agents.

Image: SiliconANGLE/Gemini