Explorations / Founding · Healthcare AI
What founding in healthcare is teaching me
I left scale to find out what I’d build with my own name on the mistake. Right now that’s a clinical AI in pilot, and the company is teaching me faster than the code. These are notes from inside it, while I still don’t know how it ends.
What it taught me
Now: in pilotThe further in I get, the more the hard part is who the system is accountable to, not what it can do.
Still open
I don’t know yet whether this becomes a company, a product inside someone else’s, or the education that tells me what to build next.
Why I’m here
What I left scale to find out
I left scale to find out what I’d build when it was my own hands on the line and my own name on the mistake. Right now that’s a clinical AI for specialist practices, in pilot. I expected to spend my time on models. I’m spending it on what it takes to be trusted with something that matters, and how little of that turns out to be the model.
Where the work lives
Care happens between the visits
The first thing the practice corrected: the care it was losing wasn’t lost in the exam room. It leaked out of the lists beside the calendar: the recall nobody booked, the prep nobody walked through, the no-show nobody had time to chase. The agent works those lists, answers the routine questions, and hands the exceptions back to staff.
The agent works the gaps; the exceptions go back to a person
Fig. 01 — The visit is watched; the care leaks from the gaps around it
watchedthe gapsWhat earns trust
The specialists kept correcting it
I came in proud of the agent’s fluency. The specialists kept correcting its judgment: what it could say, when to stop, when to hand a call to a nurse. So the work became getting their rules out of their heads and into protocols the agent runs against, and the practice can read, version, and change. When it gets a call wrong, they can ask which rule allowed it, and fix the rule.
I stopped trying to make the agent trustworthy, and started making it accountable.
Agent gets it wrong → ask which rule allowed it, fix the rule
Fig. 01 — The judgment moves from the expert's head into a rule a person owns
in their headthe protocolWhere the human stays
The agent that raises its hand
The version of this that scares me replaces the nurse without anyone deciding to. The one I’m building works for her: she triggers it, it makes the call, it writes the outcome back, and it stops the moment a case needs a person. Autonomy gets earned in inches, on evidence, away from that stop.
Triggered by a person → routine handled → the hard case returned
Fig. 01 — A person starts it; the moment it needs judgment, it hands back
the routinethe handoffWhat it’s teaching me
The part I can’t answer yet
Two things keep getting truer. The leverage and the risk both live in the unglamorous edges around the model. And the thing that decides whether any of it ships is trust: the protocols, the handoffs, the record of what the system did and why.
I still don’t know where this goes. It might become a company, a product inside someone else’s, or the education that tells me what to build next. I’m holding that open on purpose. What I won’t do is pretend I’ve decided, because the honest version of this season is that I’m still choosing.