Amazon says thousands of AI agents wrote an entire software product, with no human writing or reviewing the code. Here is how it worked and what it means for development teams.
Most people now accept that AI can help write code. Fewer expect it to build a whole product without a human touching the code.
On November 11, 2026, Amazon Web Services released AWS Emulator and made exactly that claim. According to Amazon, autonomous AI agents wrote the tool entirely, and its engineers neither wrote nor reviewed the code. The interesting part is not just the claim, but the system Amazon built to make it work.
What Is AWS Emulator?
AWS Emulator is a free tool that lets developers run AWS services on their own machine. Instead of sending requests to the real cloud, the AWS command-line tools, SDKs and coding agents talk to a local copy. It can run on a laptop, a development server or a continuous integration pipeline, without an AWS account.
It is built from 438 AWS API models and covers more than 200 services. Amazon released it as open source under the CC0 public-domain dedication.
For developers, the benefit is speed and cost. You can test code, try ideas and run automated tests without deploying cloud resources each time. It also gives AI coding agents a safe place to create resources, check results and fix their own mistakes.
How Did AI Agents Build It?
The project started as a small experiment. An AWS engineer asked Amazon’s Kiro coding agent to write tests, check them against real AWS and build a version that passed them. In about 15 minutes, it produced a working Amazon SQS emulator.
Amazon then scaled the idea into a system with three teams of agents, each with a narrow job:
- Test writers create tests from AWS’s public API models and verify each one against real AWS before saving it.
- Issue finders replay those tests against the emulator, spot where its behavior differs from real AWS and file issues.
- Fixers take one issue each, find the root cause and write the fix.
Amazon says tens of thousands of these tests now exist. Hundreds of agents work at the same time, and together they fix thousands of issues every day.
The key idea is that each agent has a well-defined task. Amazon says agents perform better when they are given specific work instead of vague goals.
The Merge Gate: How Amazon Controls Quality
If no human reviews the code, something else has to. Amazon’s answer is a merge gate, a set of automated checks every change must pass before it enters the codebase.
According to Amazon, the gate checks that the code compiles, follows coding standards, passes security reviews and does not make the emulator less accurate than before. Several security agents review each change adversarially, looking for ways it could be attacked, and a code reviewer agent checks it as well. If any check fails, the change goes back to be fixed.
After a change is merged, it moves through a release pipeline built on AWS developer tools before the finished product is published.
What Did the Humans Do?
Engineers didn’t write the emulator, but they did not disappear. They built everything around the agents: the specifications, the test machinery, the merge gates and the release pipeline.
This is the real lesson. When agents write the code, the engineer’s job shifts toward defining what “correct” means and building automated systems that enforce it. The quality of the output depends on the quality of those guardrails.
What It Means for Software Teams
Tests become more valuable. In this setup, tests verified against reality are the foundation. Agents can rewrite code endlessly, but the tests decide whether the result is right.
Small, focused agents work better. Splitting work into separate roles, such as writing, checking and fixing, kept each agent’s job clear and made the system easier to control.
Automated checks are essential. Security review, coding standards and regression tests must run on every change, especially when changes arrive at high speed.
Human judgment still matters. People decide what to build, how to measure success and what the guardrails should be.
Limits and Cautions
Before treating this as a blueprint for every project, keep a few points in mind.
Much of what we know comes from Amazon’s own description. Independent verification is still limited.
The emulator is also not a perfect copy of AWS. It does not evaluate access policies the way real AWS does, so a request that AWS would deny could succeed locally. It is meant for building and testing, not for production workloads or security testing, and gaps in compatibility with real AWS remain.
Finally, Amazon had unusual advantages: a clear specification in the form of public API models, and a real AWS environment to check every test against. Many business projects don’t have a reference that clear.
Building Real Products Still Needs Experienced Teams
AI agents are getting better at writing code, but a customer-facing app needs far more than code. It needs clear requirements, good design, security, performance testing, a smooth user experience and long-term maintenance.
That is where experience matters. If you are planning a new product, working with a trusted app development company in USA can help you combine modern AI tools with solid architecture, testing and security practices, so you get speed without losing quality.
Final Thoughts
Amazon’s AWS Emulator shows what becomes possible when AI agents are given focused roles, strong tests and strict automated checks. Whether or not every team can copy the approach, the direction is clear: software development is moving toward people designing the rules and AI doing more of the building.
For businesses and developers, the takeaway is practical. Invest in clear specifications, good tests and automated quality checks. They will matter whether the code is written by people, by AI or by both.
