SDSignal Desk

AWS Releases AWS Emulator: A Local Copy of 200+ AWS Services Built by AI Agents

· Signal Desk newsroom

Analysis of reporting by MarkTechPost. How we work

Image: MarkTechPost

AWS let fleets of AI agents write an entire local emulator of its cloud, and the honest fidelity scores show both how far agentic coding has come and where it still falls short.

Why it matters

AWS Emulator is a free local stand-in for AWS APIs that runs on a laptop, a dev server or in CI, with no AWS account and no service fee. It covers more than 200 services built from 438 API models, under a CC0 license. Teams can run fast, cheap tests against AWS-shaped endpoints without touching the real cloud.

The bigger news is how it was made. MarkTechPost reports that autonomous agents wrote and reviewed all of the code, with humans designing the tests and merge gates. This is one of the clearest public examples of agents building a large production tool end to end.

From the desk

We like this a lot, and mostly because of the transparency. AWS publishes fidelity scores with every release, including the failing tests. At version 1.0.322, the emulator passes 24,243 of 42,857 scored tests, or 56.6 percent. Some services are strong: KMS near 94 percent, SQS near 90, S3 at 79. CloudFormation sits at about 16 percent. That is not a polished launch number. It is an honest one, and it gives developers what they need to decide where to trust it.

The build process is the real lesson. Agent teams split into test writers that derive scenarios from public API models and check them against real AWS, issue finders that replay those tests, and fixers that each take one issue. Every patch passes a merge gate for build, code review, security and fidelity. That is the right shape for agentic software: humans own the definition of correct, and agents grind through the work. It suggests the bottleneck in AI coding is shifting from writing code to writing good tests.

The risks are clear, and AWS states them plainly. There is no IAM policy evaluation, so it should never be used for security testing. It runs one region and one account, in memory, and is not built for load. Telemetry is on by default, which some teams will want to switch off. The danger is a team seeing green tests locally and assuming production will behave the same, especially on lower-fidelity services.

There is also a market angle. Paid tools like LocalStack charge for broader coverage. A free, first-party emulator with this breadth puts real pressure on them. Good for developers in the short run, harder for the independent ecosystem that filled this gap for years.

Context

MarkTechPost reports the project began with a prompt to AWS's Kiro agent that produced a passable SQS emulator in about 15 minutes. Releases moved from v1.0.265 to v1.0.322 between November 8 and 11. Source builds need about 100 GB of disk; packaged versions run on Linux or in Docker on macOS and Windows.

Who feels it

Developers
A free local loop for many AWS services speeds testing, as long as teams check per-service fidelity scores first.
Coding agents
An installable skill lets agents test against AWS-shaped APIs without cloud costs or credentials.
Security teams
No IAM evaluation means passing tests say nothing about permissions; real-cloud security testing remains essential.
Emulator vendors
LocalStack and similar tools face a free, broad, first-party competitor.

What to watch

  1. How fast the overall fidelity score climbs with each release
  2. Progress on weak areas like CloudFormation, Lambda and EC2
  3. Whether other cloud providers copy the agent-built, test-gated approach
  4. Pricing or feature responses from LocalStack and other emulators

Read the original

Continue at the source.

MarkTechPost

More on AI

All AI coverage →