AWS Ships an Emulator for Its Entire Cloud Built Entirely by AI Agents, With No Human Code Review

Amazon released AWS Emulator on Wednesday, a free tool that simulates more than 200 of its cloud services on a developer's own machine — and the company is making a far bolder claim about how it was built. According to AWS, the emulator was written entirely by autonomous AI agents. Its engineering team, AWS says, has not written or reviewed a single line of the code.

The tool lets developers point the AWS CLI or any official SDK at a local endpoint that behaves like the real cloud: calls return in milliseconds, no AWS account is required, and each copy keeps its own independent state. Under the hood it is built from more than 400 API models covering services spanning compute, storage, queues, IAM and telemetry. AWS has open-sourced the project on GitHub under a CC0 license.

Senior principal engineer Matt Goldberg, who announced the release, described the origin as a weekend experiment. Inspired by Anthropic's earlier demonstration that agents could autonomously build a C compiler given a spec and a reference implementation, he gave the Kiro coding agent a three-step prompt: write tests from the AWS CLI, iterate until they pass against real AWS, then build an in-memory implementation those tests pass against. Fifteen minutes later he had a passable first version of Amazon SQS. By the end of the weekend, Kiro had produced a rough version of the rest of AWS, and his team started testing its own software against it the following Monday.

Turning that experiment into a product took an agent factory. AWS runs three teams of agents on Amazon Bedrock AgentCore: one writes fidelity tests against the real cloud, one replays those tests against the emulator and files issues for every gap, and a third fixes them, one agent per issue. Every change passes through an automated merge gate that checks coding standards and security. Goldberg says hundreds of agents now fix thousands of issues a day, and the same loop keeps the emulator in sync as real AWS evolves.

The release lands in a crowded niche. Projects like LocalStack and the Moto library have offered fake AWS environments for years, and the Testcontainers movement made disposable infrastructure in CI mainstream. What is new is that the cloud vendor itself is shipping the emulator, free and first-party — an implicit acknowledgment that customers want the cloud's API surface without the cloud's logistics, and a direct challenge to paid emulation tools that have owned this space for a decade.

But the provenance claim is the part the industry will argue about. A company known for operational paranoia has shipped production software with no human code review, with humans reduced to writing specifications and designing the guardrails that keep agents inside them. If the approach holds, review bandwidth — not typing speed — becomes the constraint on engineering throughput, and the junior-engineer on-ramp of learning by writing low-stakes code erodes further. The caveats are real: an emulator is not production AWS, fidelity gaps remain, and the 100-percent-agent claim is AWS's own description rather than an independently audited fact. Still, as a named, open-source data point in the debate over whether agents can build real software, it will be cited for a long time.