Architecture

What I write about here ​

Most posts under this tag share a thesis. The bottleneck on AI productivity is almost never the model. It is the system the model sits inside.

A coding agent that can write a function is useful. A coding agent that can write a function and push it to production without anyone reviewing it is a security incident waiting to happen. A pipeline with no integration tests turns every AI commit into a coin flip. None of that is the model's fault, and none of it is solved by upgrading to the next one.

So the architecture work I write about is the work around the agent. How wide the blast radius is when something goes wrong, where the boundary between intent and execution sits, what you put between a model's confident output and your actual production database.

The Laravel post is one practical pattern. The Gemini integration is another. The database deletion story is what happens when those boundaries are too thin.

If you came here looking for clean diagrams of microservices, you are in the wrong place. I write about architecture as containment, mostly. What the system protects you from, and what it cannot.