After a year of Claude Code in my daily workflow, here's the honest version of how AI fits into real engineering work — the tasks I hand it without hesitation, and the ones I'll never let it own.

I built a Retrieval-Augmented Generation (RAG) app that answers questions from your documents — and shows the exact passages behind every answer, with match scores. Here's how it works and why retrieval, not the model, is the hard part.

Most codebases have no tests because writing them is annoying, not because engineers can't. Here's how I used Claude Code and Cursor to go from near-zero to 70%+ coverage — without letting AI replace my judgment.

Stripe charged the customer once — but my system recorded it three times. The cause wasn't a bug in my logic. It's that event systems deliver "at least once," not "exactly once." Here's how idempotency fixes it.

Every endpoint was fast, but screens still felt slow. The culprit was over-fetching. Here's how I used Claude Code and Codex to trace it and cut payload size by ~30% — without handing over the decisions.

Five years moving across the stack, here is how I think about whether a role is a Full Stack Developer, Frontend Developer, or Backend Developer fit.

A practical setup for knowing what users actually do, catching errors before users report them, and trusting your own deploys.

How offloading GHIN/USGA handicap score calculations to Celery + Redis kept a Django REST API responsive under load.

What it actually took to ship a new flow, a full UI rebrand, and a chatbot integration on a production Next.js + GraphQL platform with active users.