Back to writing
February 20267 min read
Building PebbleCode on AWS
Notes on why the platform ended up serverless, where the architecture bends, and what I would simplify next.
ArchitectureStartupsAI Systems
PebbleCode runs on a serverless AWS stack because it matched the shape of the product early on: bursty usage, small surfaces, and a need to move quickly.
Cognito, API Gateway, Lambda, DynamoDB, and S3 let me build a full product loop without managing too much infrastructure overhead. That said, every convenience creates a different kind of complexity in observability and local debugging.
The tradeoff was acceptable because the product needed speed of iteration more than infrastructure purity. If the system grows in a certain direction, I will probably replace pieces of it later.