The Future of Healthcare UX

Sebastian Caliri and Finn Kennedy at 8VC published "A Vision for Healthcare AI in America," and it lays out the clearest roadmap I've seen for how healthcare AI will actually become legal, billable, and scalable.

Level 0: Administrative AI. Scribes, scheduling, prior auth automation. Already happening. Already legal.

Level 1: Assistive AI. Coaching, education, care navigation. The AI helps, but doesn't diagnose or prescribe. Currently blocked by reimbursement.

Level 2: Supervised Autonomous AI. AI that can diagnose and prescribe, but under physician oversight. Currently blocked by FDA clarity, state medical practice acts, and reimbursement.

Level 3: Fully Autonomous AI. The "AI doctor." Currently illegal under the Social Security Act.

Most healthcare AI companies are building Level 1 and Level 2 products while the regulatory infrastructure is stuck at Level 0.

As AI moves from Level 0 to Level 2, the trust requirements compound.

At Level 0, AI is invisible. The scribe transcribes. Nobody needs to trust the AI because nobody's making clinical decisions based on it.

At Level 1, AI is advisory. Trust matters, but the human is still fully in control.

At Level 2, AI is making decisions that physicians are signing off on. Now trust is load-bearing. If the physician can't understand WHY the AI recommended something, they won't approve it. If the patient can't understand WHY their medication changed, they won't comply.

Remember Edition 2? Defensible over intuitive?

That principle becomes existential at Level 2. Every AI recommendation needs to be explainable. Not to satisfy regulators. To satisfy the physician who's putting their license on the line.

Here's what the policy wonks miss:

Every level of this framework assumes the AI has unified patient context.

It won't.

Not when medication history lives in three systems that don't talk to each other. Not when the GP record doesn't know what the specialist prescribed. Not when legacy EMRs export data in formats that modern AI can't consume.

The AI works. The regulations are (slowly) evolving. But the interface layer that connects patients, providers, and AI systems? That's where everything breaks.

Level 2 and Level 3 AI won't fail because the models aren't smart enough. They'll fail because the interfaces don't surface the right information at the right moment to the right person.

The interfaces that win at Level 2 will be the ones that make AI reasoning visible, auditable, and defensible.

The ones that hide AI reasoning behind "trust the algorithm" will die at 15% adoption, just like every other healthcare product that forgot clinicians need to explain their decisions to supervisors.

This newsletter is ending. But the conversation isn't.

The AI Trust Gap series will cover:

• Why "AI-powered" became a liability, not a feature
• The transparency spectrum: when to show AI reasoning, when to hide it
• Trust calibration: why users need to trust AI the right amount
• The explainability trap: why "explainable AI" isn't enough
• Standards that are emerging
• Case studies of trust done right

Whatever you're building, whatever legacy system you're staring at, whatever adoption problem is keeping you up at night: you're not alone. The pattern exists. The solutions exist. The hard part is having the courage to see the real problem.