AWS Emulator and the engineer's job when agents write the code
AWS has released AWS Emulator, a free offline copy of AWS that coding agents can test against, and the whole thing was written by agents checking themselves against the real cloud. The piece also argues that the engineer's job shifts toward writing specs, tests, and merge gates that decide when agent-written code is correct.
- The emulator covers more than 200 AWS services and was built from 400+ public API models in Smithy, with every line written and tested by agents.
- Agents work from fidelity tests that record how real AWS behaves, then build the emulator until it passes the same tests; a prototype SQS emulator took about 15 minutes from one prompt.
- The author's own experience: an agent struggled for four nights rebuilding a replication engine with no way to check its results, then finished in about two hours once it had a spec and a test environment.
Swami Sivasubramanian's story anchors the piece. In 2004, after a bad Christmas when commercial relational databases couldn't keep up with Amazon's load, Werner Vogels had Swami dig into real database usage. He found 70% of operations were key-value, not relational, and that finding led to Dynamo. Later, Swami and Al Vermeulen built Alf, a distributed lock service on Paxos that now sits underneath systems like S3 and DynamoDB. The hard part was testing it. Chaos testing and large-scale runs were expensive and couldn't reproduce many failure orderings on demand, so the team ran the servers inside a single JVM with a simulated network, injected packet loss and faults at will, and used TLA+ for formal checks. That caught most concurrency bugs before real deployment, though real hardware still surfaced problems the simulation missed.
The new problem is scale and speed. AWS now has over 200 services, and every application that calls them needs a test environment. Hand-building those would take years, and agents produce code faster than teams can validate it in production. The approach Matt Goldberg proposed uses the Smithy API models as the written spec and the live AWS cloud as the reference. Agents turn public API behavior into fidelity tests, such as checking that an SQS message stays hidden until its visibility timeout expires, then tune those tests until they pass against real AWS. Agents then build an emulator that must pass the same tests, and any failure points to a gap. A first SQS prototype came out of Kiro in about 15 minutes from a single prompt. AWS Emulator, released today, is the result: more than 400 API models, over 200 services, all code written and tested by agents, running free and offline on a laptop.
The practical upshot for engineers is in how the work shifts. A coding agent works in a write-run-read loop, and when the services it calls behave like the real ones, it can check its own output as it goes and get things right on the first try more often. So the engineer's job moves toward writing the specs, the fidelity tests, and the merge gates that decide whether agent output is correct. The piece notes that every change to the emulator passes a deep set of automated checks, including security agents doing adversarial review. The author admits he learned this the hard way while using Kiro to rebuild a replication engine, which is a useful reminder that the gates matter more than the generation.