AWS Built Its Own Emulator: What It Means for Local Development and Production AWS released AWS Emulator on 11 November 2026, a free Rust-based program that answers AWS API calls on a developer's own machine and is released under the CC0 public domain dedication. Matt Goldberg, a senior principal engineer at AWS, wrote in the launch story that "The emulator was written entirely by autonomous agents. Our team hasn't written or reviewed any of the code," with three teams of agents maintaining it through fidelity tests scored per service against published benchmarks. AWS positions the tool for coding agents and CI jobs, giving each emulator instance its own port, keys and state, but the company notes state lives in memory and is lost when the emulator stops, and that it matches control plane APIs most closely while data plane behavior varies by service. On 11 November 2026 AWS released AWS Emulator, a free program that answers AWS API calls on your own machine. AWS says agents wrote every line of it. Both facts tell you where cloud development is heading, and both raise the question this post answers: once your AWS code runs locally for testing, where does it run for real? AWS heard the demand for local AWS Developers have run AWS locally for years. Moto has mocked AWS in Python tests since 2013, LocalStack built a company on it, and Floci passed 30,000 GitHub stars in its first nine months. The appeal is simple: tests that run in milliseconds, cost nothing, and need no account or network. Coding agents have raised the stakes. An agent that writes AWS code needs somewhere to run it, hundreds of times an hour, without a bill or the risk of touching a real account. AWS built for that directly. Each emulator instance gets its own port, keys and state, so parallel agents and CI jobs don't collide, and AWS ships an aws-emulator skill that a coding agent installs with one AWS CLI command. When the company that sells the cloud ships a free tool for building without it, local emulation has become a standard part of how people build on AWS. Built entirely by agents The second signal is how AWS built it. Matt Goldberg, a senior principal engineer at AWS, wrote in the launch story https://aboutamazon.com/news/aws/how-autonomous-ai-agents-built-aws-emulator : "The emulator was written entirely by autonomous agents. Our team hasn't written or reviewed any of the code." It started with Goldberg asking Kiro, AWS's coding agent, for an SQS emulator, and getting a working first version 15 minutes later. The product is now maintained by three teams of agents. One writes fidelity tests from AWS's API models and checks them against real AWS, one runs those tests and files every difference, and one fixes them, one issue at a time, through a merge gate that includes adversarial security reviews. AWS's leadership is framing this as the future of software, not a stunt. Writing on Werner Vogels' blog, Swami Sivasubramanian put it plainly: " We expect agents to build and check most software https://www.allthingsdistributed.com/2026/11/aws-emulator-and-the-engineers-job-when-agents-write-the-code.html , with engineers deciding what gets built and when the work is correct." When AWS ships a product that no human wrote, enterprises take note of how fast their own teams will be expected to ship. What emulation is, and what it isn't An emulator answers AWS API calls with responses in the right shape. Your SDK, CLI or Terraform sends the request it would send to AWS, and gets back what AWS would return. Create a bucket, put an object, read it back, and your test passes on a laptop with no connection. Emulators differ in what sits behind that response. Some keep a record of the resource and nothing more. Others run the real engine: a Postgres container behind an RDS call, or your Lambda function executing in a container. The first checks that you called the API correctly. The second checks that your code works. No emulator gives you the parts of AWS that make it a cloud: virtual machines on real hardware, durable replicated storage, networks isolated between tenants, and IAM checking every request against your policies. Those are what production runs on. Where AWS Emulator fits AWS Emulator generates its services from the API models the AWS SDKs are built from, so new AWS services appear in it quickly. It's written in Rust, released under the CC0 public domain dedication, and scored per service against published fidelity tests https://docs.aws.amazon.com/emulator/latest/userguide/fidelity.html . AWS is clear about its limits: - It matches control plane APIs most closely, and data plane behaviour varies by service. - State lives in memory and is gone when the emulator stops. - It doesn't evaluate IAM policies, and AWS says not to use it for security testing or production. - Telemetry is on by default, though you can turn it off. - The repository takes bug reports, not pull requests. That makes it a strong fit for fast unit tests and for coding agents working against AWS, which is what AWS designed it for. Teams that want more than API shape have options. Floci, MIT-licensed and community-run, runs real Docker containers behind Lambda, RDS, ECS and EKS, can keep state between runs, can enforce IAM in CI, and sends no telemetry. It also takes pull requests, so a gap you hit is a gap you can fix yourself. LocalStack, the longest-running option, has required an account and auth token since it archived its Community Edition in March 2026. From local emulation to local production AWS Emulator answers the first half of a question many teams are asking: can we build and test AWS workloads without AWS? The second half is whether those workloads can then run without AWS, in your own data centre, at the edge or in an air-gapped facility. An emulator can't answer that, and AWS tells you not to try. Spinifex is built for that second half. It's an AWS-compatible cloud you run on your own servers, and it's not an emulator. When you call RunInstances on Spinifex, a KVM virtual machine boots from your AMI. EBS volumes are replicated block devices, S3 objects are erasure-coded across nodes, security groups filter real traffic, and IAM evaluates every request. Your resources persist until you delete them, and a three-node cluster keeps running when a server fails. The workflow joins up through the AWS API. Your developers and their coding agents build against an emulator on the laptop and in CI. The same code, the same Terraform and the same SDK calls then deploy to Spinifex, with only the endpoint and credentials changed. AWS Emulator makes the first step easier for everyone, and Spinifex picks up the workload when it's ready to run in production on hardware you control. Get started AWS Emulator is at github.com/aws/aws-emulator https://github.com/aws/aws-emulator , and Floci is at github.com/floci-io/floci https://github.com/floci-io/floci . Spinifex is at github.com/mulgadc/spinifex https://github.com/mulgadc/spinifex , and its AWS API coverage https://docs.mulgadc.com/coverage page lists the services it serves today. For the full path from emulator to production, read what changes when your AWS code runs for real https://mulgadc.com/blog/localstack-to-production . To try Spinifex without hardware, the free Spinifex sandbox https://mulgadc.com/signup gives you a live endpoint for 72 hours.