About
I'm Jon Roosevelt — I've built a fleet of AI agents that runs a growing share of my company's work and my own, and I share candidly how it actually works.
I'm Jon Roosevelt. I've built a fleet of AI agents that now does a real, growing share of my own work and my company's operations. I've been programming since 2007, and I build these systems end to end — the agents, the memory and orchestration behind them, and the infrastructure that makes them dependable enough to trust with real work. I've aimed all of it at healthcare, because that's where I think the impact is largest — and it's personal.
Why I do this
Remember chopping vegetables when the knife slips? My wife — a violinist — almost lost her index finger in a kitchen accident. We rushed to the local urgent care around 7 p.m. To our surprise, there were no other patients. You'd think we'd be seen immediately. Instead: "You'll be called when the provider is ready." I held her while we sat through an agonizing 45 minutes, both of us wondering if she'd ever play on stage again.
The provider, when we finally saw one, was exceptional — the cut was cleanly stitched and she was back on stage in a few weeks. But on the drive home, the idea for Arcs Health took root. An empty waiting room and a 45-minute wait shouldn't coexist. I started imagining clinics where waiting times are effectively zero, where providers spend their time with patients instead of drowning in busy-work — no more "pajama-time" charting at midnight, no more prior-authorization headaches, no pressure to cram more visits into a day.
What I'm building
I run Arcs Health, where I'm using modern AI to take the friction out of urgent care — ambient charting, smarter scheduling, automated insurance and intake work, a patient portal, and an agentic harness that runs it all — so clinicians can practice medicine instead of fighting software. That work runs in real clinics, including Covenant Clinics, where the systems I build meet messy, high-stakes reality every single day.
Urgent care is a perfect proving ground for AI: high volume, time pressure, real consequences, and an enormous amount of work that has nothing to do with patient care. If you can make an agent dependable there, you've learned something that holds up anywhere.
How I build
I work across the whole stack of agentic systems — the agents themselves, the RAG and memory that ground them, the orchestration that coordinates them, and the infrastructure that makes them dependable. I'm less interested in demos than in systems that hold up against reality: the boring, load-bearing parts — verification, observability, safe deployment, the failure modes nobody screenshots. The tools I reach for, today:
Curiosity, broadly
My interests run wider than code. I read across philosophy, psychology, leadership, and the history of how people actually get things done — and a lot of it ends up shaping how I build. You'll find that range on the blog: agentic engineering and urgent-care operations next to notes on motivation, meaning, and decision-making. I'm a generalist who likes going deep — curious about almost everything, and happiest building at the edge of it.
This site
It's where I share my work openly — so you can see not just what I built, but how. The blog is for essays and field notes I figured out the hard way; the garden is rougher, living notes I tend over time; the now page is what has my attention this week; projects is what I've built. Everything technical is generalized — principles and lessons, never anything specific to a patient or client.