Deploying an app to Kubernetes on AWS, with AI as my pair engineer
Before AI, you'd settle on one infrastructure design up front and then build it. Today, AI lets me compare several approaches quickly and pick the most reliable one for the job. The design phase moves much faster. It's still iterative, though: the infrastructure gets tuned to the real use case one step at a time.
To try this, I used Claude to take a simple Spring Boot app all the way to Amazon EKS (Kubernetes on AWS). The app wasn't the point. The point was the process of getting code into production reliably.
What the process covers:
- Build and test the app locally in Docker
- Push the image to Amazon ECR
- Create the Kubernetes cluster
- Deploy the app behind a load balancer
- Release a new version with zero downtime
- Tear everything down so nothing keeps costing money
It wasn't a one-shot build. Getting it smooth took several rounds of
- AWS rejected the server type on a new account's free plan
- An image built on a Mac wouldn't run on AWS servers until it was cleaned up from previous run
- Version tags got out of step, so the cluster kept looking for an image that didn't exist
Every fix went back into the scripts and the documentation, so the next run just works.
The result is a "first cut": one small script per step, run in order, repeatable by anyone following the README.
Next phase: full automation with a CI/CD pipeline, so every code push builds, tests and deploys the new version on its own.
Code and step-by-step guide: https://lnkd.in/g_GNvWqf
Top comments (0)