I don’t start with features. I start with the workflow someone already has, badly.

Before HealthFirstAI, I would have said the job of a product is to give people a better tool. Now I’d say the job is narrower than that: make an existing behaviour faster, or make an avoided behaviour possible.

Doctors weren’t avoiding documentation because they didn’t have a tool. They were avoiding it because every tool asked them to stop treating the patient and go fill a form. The behaviour I needed wasn’t “use this software instead.” It was “keep doing what you’re already doing, and let the software catch up.”

That’s a smaller ambition than it sounds. It’s also the only one that survives contact with a real hospital.