How I Interact with AI

I’m aware that coming into a field with significant experience can be a hindrance when there are revolutionary new tools, so my goal with Claude Code has been to embrace an AI-native approach as fully as possible. My previous interactions with AI were autocomplete-on-steroids. Lately I’ve tried to let the machine write as much of the code as possible.

My first attempts had the machine writing close to 100% of the code, with me as an incredibly pedantic reviewer. At first it was great—I would never critique a human as closely as I did the machine. More recently, I’ve actively tried to loosen the reins. So long as the machine achieves its end goals, I try to avoid pushing it too hard to match my style.

Workflow

• A GitHub-driven workflow: define issues, open PRs, review
• CI checks to enforce best practices deterministically: linting, formatting, testing

An LLM has no taste, and can take no responsibility. The first one I’m finding over and over.

My role is shifting from coder to designer. I spend more time on UX decisions than implementation details.