How one physician — told it would take an agency and a year — built the AI that now runs his practice. In four months. Alone, with a frontier model.
An enormous CRM to run the office — roughly 800 properties, 381 workflows, ten pipelines. On paper, a machine.
365 of the fields I’d built were never touched. The system had outrun everyone in the building — and the work it was supposed to do simply wasn’t getting done.
I’d decided this was the year of AI agents. I’d already shut off one vendor’s agent — it was no good — and I handed the agency a roadmap with one thing at the top: deploy AI agents for SMS. Then I waited.
I stopped waiting and built it myself, in Claude Code. On February 14, 2026, the SMS agent went live — booking real patients, in English and Spanish — two days before I’d even opened the repository that became all of this.
Eleven systems, all live in my practice today — built between February 16 and June 16, 2026. Open any one for its own page →
Texts patients back any time — day or night, in English or Spanish — and books their appointments for them.
The one app my whole office runs on — scheduling, patient records, billing, and the front desk, all in one place.
Around a hundred behind-the-scenes helpers that quietly handle the repetitive office work, around the clock.
Our online ads had quietly stopped working. I found out why and rebuilt them so they bring patients in again.
Listens to every phone call and writes a clear summary into the patient’s file within minutes.
The wiring that links the AI to our medical-records system, so it can book real appointments — not just pretend to.
Records a visit on an iPad and writes up the doctor’s note automatically, in English or Spanish.
The tools I built to double-check my own work, so the AI can move fast without making mistakes.
Snap a photo of a paper patient form and the right record is created for you in about thirty seconds.
Makes sure “I reached the patient” really happened — it checks the phone and the calendar instead of taking someone’s word.
A Mac app for the doctor: hold a key, talk, and your words turn into clean text anywhere on screen.
Anyone can call a frontier model. Shipping safe, production systems as one person comes down to a handful of disciplines — the ones I’d set up for any practice I work with.
I fact-check the model against code, version history, and live systems — never the prose. This very site was put through an adversarial review and corrected where it was wrong.
Structure at every boundary, so a model physically can’t leak its reasoning — or a wrong date — into a patient’s text. Correctness is enforced, not requested.
Every risky feature ships behind a flag and a safe default. Nothing I deploy is irreversible, so I can move fast without betting the practice on it.
Automations are generated from versioned scripts, not hand-edited. The system can always be rebuilt, diffed, and audited — which is how one person safely runs ~97 of them.
I’m a practicing physician who runs my own practice and builds its AI myself — not an agency, and not rented software. Here’s how we could work together.
A clear, ranked plan of what AI can actually do for your practice — and exactly what to build first. A fixed-scope place to start.
I build one of these into your practice, end to end — the patient texter, the scribe, call intelligence — live and working, not a demo.
The full build: the app your office runs on, the automations behind it, and the wiring that ties it together — owned by you.
I stay on as your fractional AI lead — building, watching, and improving the systems as your practice grows.
Everything is built in-house and owned by you — your data stays yours, every risky feature has an off-switch, and nothing is rented from an agency.