Ask an AI to build you an app and you'll get a React front end and a Node.js with Express back end. Ask again tomorrow, for a different app in a different industry, and you'll get the same thing. Models will usually suggest the most common answer in their training data, and the internet contains a lot of React and Node.js examples.

If you're not careful, this can cause every problem to be solved with the tools. A decision your team used to argue about at a whiteboard now gets made by token frequency before anyone has a chance to weigh in.

I've spent years helping teams pick technology that fits their needs and constraints, and AI hasn't ended that work. It has just moved it earlier, because a model will happily honor a good architecture decision, but only if somebody makes one.

In this talk we'll make architecture an explicit step again: choosing your stack before code generation starts, writing decisions down where agents actually honor them (CLAUDE.md, ADRs, custom skills/instructions), and using architecture tests to catch drift in your CI process.