From admin checklist to point of care assistant

A clinical assistant used by primary care physicians had grown into four disconnected experiences and feature bloat. It was being used as an end-of-visit administrative checklist, which was neither what it was for nor where its value was.

The product served physicians seeing 40+ patients a day in a value-based care model, where the quality of the encounter drives both outcomes and reimbursement. It had accumulated real capability — a scribe, a patient-data chat, a visit briefing, a task system — and each one worked. The problem was that none of them knew about the others, and the center of gravity had drifted to after the visit: a checklist you cleared once the patient had left.

The AI capabilities arriving over the next year would not fit into that structure at all. The transformation had to be defined end-to-end based on workflows.

FORCING THE HARD QUESTIONS

I built the workshop output into an end-to-end vision prototype: the whole visit, pre-visit through post-visit, as one continuous experience. A prototype you can move through forces the hard questions into the open; what happens between steps, what the physician sees when the ambient layer is listening, what the agenda looks like when a patient has nothing going on?

The flow ran in four stages — Schedule, Visit, Visit + Task, Visit completion — drawn at three breakpoints. Drawing the small breakpoint first, and completely, is what made mobile-first a constraint the teams designed against. Starting with the assumption that the provider would start with the agenda, and move into ambient scribing to focus on the patient, and have the visit note ready by the end of the visit. Each stage carried a user story, a hypothesis stated so it could be tested, open questions written into the artifact, and an explicit Included / Not Included boundary — so the scoping conversation happened inside the design, at the moment the tradeoff became legible.

Then the vision became an operating plan. Working with product leadership I broke it into workstreams sized by unknown, sequenced them across two quarters, and helped form pods around workflows. I held the experience architecture while a new design system was built underneath it and each team pushed its piece forward against real engineering constraints.

• SIZED BY UNKNOWN

  Each workstream was assessed on how much we did not know and how much impact it carried. The task system carried the highest uncertainty of anything on the list, so it was explicitly deferred to prevent blocking execution. The clinical agenda, high on both axes, got prototyping capacity early and a deliberately small validation target: one deeply engaged physician using it for real visits by end of quarter.

• TEAMS AROUND WORKFLOWS

  Workflow-owned teams argue about the seams, which is where the experience actually lives. A weekly working group kept the pods composable; a monthly cross-pillar session with the CPO and clinical lead kept the whole thing pointed at the vision.

• DISPOSABLE CODE AS A DESIGN STRATEGY

  Work ran in a feature-flagged module sandbox with its own data layer, on the explicit agreement that the code was throwaway. That removed the fear that makes exploration cautious, and let eight workstreams run in parallel without paralysis about committing to the wrong thing.

• Convergence is a different skill from exploration

  By midsummer the pods had produced roughly fourteen prototypes of the live visit experience. This was an intentional part of the strategy which allowed the team diverge and explore solutions with their own unique knowlege set. I spent the back half of the summer unifying the design system across surfaces, making the layout genuinely responsive, reconciling interaction patterns where two teams had solved the same problem differently, and preserving what was good in each rather than picking a winner.

HOW WOULD IT WORK END-TO-END?

A lo-fi interactive model settled the structural questions: what opens the canvas, where Sign lives, and what the drawer is allowed to cover. Scroll through its states.

FINAL UI EMBEDDED SHOWN EMBEDDED IN AN EHR