Philosophy

What I've learned building AI in healthcare

Ten principles from 14 years working on clinical products and AI systems in regulated environments. Each one is something I require from the teams I lead — not just something I think about.

  1. 01

    Evaluate workflows before models.

    A model that performs in evaluation still has to fit the workflow. Most AI products fail at the integration point, not the inference point.

    Transcarent care navigation →
  2. 02

    Human override is a product feature.

    The ability to override, escalate, or send back for review is not a fallback — it's a core capability that determines whether the system deploys.

    Authorization states →
  3. 03

    Provenance beats confidence scores.

    Knowing where a recommendation came from — and what evidence backed it — matters more than a confidence number nobody can act on.

    Refract →
  4. 04

    Shipping safely beats shipping first.

    In healthcare, a premature launch creates downstream costs that dwarf the speed advantage. The unhappy path gets the same engineering as the happy path.

    Epic escalation process →
  5. 05

    Enterprise trust is earned operationally.

    Hospital IT, legal, and compliance teams don't trust models — they trust the operational system surrounding the model. Compliance documentation is a product decision.

    Doximity Dialer →
  6. 06

    AI products fail more often in integration than inference.

    The hard work is fitting into EHR systems, clinical workflows, and regulatory frameworks — not achieving better model accuracy.

    Case studies →
  7. 07

    The decision the AI feeds into matters more than the AI's output.

    When a system harms someone, the question is never 'was the algorithm wrong?' It's 'who owned the decision the algorithm fed into?'

    The Crumple Zone →
  8. 08

    Governance is a product capability, not a compliance exercise.

    Evaluation, monitoring, escalation, and correction should be built into the product — not bolted on after a compliance review.

    Ethotechnics →
  9. 09

    Correction propagation is the real test of a deployed system.

    When the evidence changes, the correction has to reach everything the error touched. If it doesn't, the system was never really deployed — it was just running.

    Reconsideration loop →
  10. 10

    The unhappy path deserves the same engineering as the happy path.

    Defaults before the edge case arrives. Every trigger has a default action, an owner, and a required record — defined before production, not after an incident.

    Escalation matrix →

The artifacts below are what I actually use to run product teams — not templates, but the frameworks and checklists that have survived real use.

If your organization is building healthcare AI products that must perform in real-world environments — where adoption, governance, evaluation, and measurable impact matter as much as technical innovation — I'd welcome the opportunity to help lead that effort.