Hey, I'm Jaime
Self-taught developer in El Paso, Texas. I build websites for a living and I practice agentic engineering in public — which mostly means publishing the numbers instead of describing them.
The short version
No degree, no bootcamp. Over a decade of building websites, most of it learned by shipping things and reading the error afterwards. I've called El Paso home for more than thirty years.
The day job is 915website.com — freelance web development for small businesses here in the borderland, run as a Texas LLC. If you're looking to hire someone for a site, that's the door.
This site is the other half. It's where I work out loud about running LLM agents as a real part of a development practice, and where I keep the evidence that it does or doesn't work.
Real figures from my own Claude Code database, not estimates. They come from the same snapshot that builds the receipts, and they change when it does.
How the work actually runs
Not "I use AI." Everyone uses AI. These are the three things that make it a practice rather than a party trick.
Written rules, not prompting harder
Every project carries a CLAUDE.md — banned packages, database permissions, commit conventions, the traps that cost a day once. The agent reads it at the start of every session. Rules in a file are version-controlled and auditable; rules in a prompt evaporate.
Autonomy where mistakes are cheap
On a feature branch the agent commits freely and moves fast. On main or develop it is locked and has to ask. Nothing is ever pushed or merged without me doing it by hand in lazygit.
The receipts are public
This site publishes its own usage data, its own accessibility scores, and a changelog entry for every commit — including the ones where I got it wrong. A practice that only shows its wins is marketing.
Where the evidence is
Nothing on this page needs taking on faith. Each of these is the underlying data.