7 Comments
User's avatar
Brittany Crosby's avatar

Thank you for highlighting Halle's post and sharing your proposed ways to tackle these core problems. As a product manager who has worked on Big Tech, Pharma, and for EHR companies, I've remained passionate about health literacy & behavioral change-- what good is all the data if patients aren't empowered to do something with it?

A study published by Microsoft researchers this month found that health queries is the number 1 topic for copilot users on mobile (assuming personal use) and people are seeking advice and guidance, not just basic information. This reinforced a key message for me: If we don't provide medical grade tools for patients soon, they will find their own ways with what's available to them. In an age of misinformation, this is too big of a risk.

Study:

https://microsoft.ai/wp-content/uploads/2025/12/What_people_do_with_Copilot-8.pdf

Eugene Vestel's avatar

Brittany, this is exactly the right framing.

Health data isn’t the problem. Power and agency are. If patients can’t understand or act on what we generate, then we’re just building more sophisticated systems that talk past the people who matter most.

That Microsoft stat should make everyone uncomfortable. Patients are already using general copilots for real medical decisions because the healthcare system hasn’t given them a credible alternative. This isn’t a future risk. It’s already happening.

Here’s the part we keep avoiding. If we build provider-only copilots that live inside closed EHR ecosystems and can’t share context, reasoning, or evidence with patients, we’re recreating the same failure mode healthcare has always had. We make clinicians faster, but we leave patients dependent and confused. That is not patient-centered care. It’s just more efficient opacity.

I think there’s a real collision coming. Closed, proprietary copilots optimized for billing, compliance, and monetization versus patient-facing intelligence built around evidence, transparency, and lived context. If those systems can’t interoperate, the value breaks. A clinician having “the answer” in a black box that a patient can’t see or carry forward is not progress.

On monetization, you’re right. Healthcare will absolutely try to gate and price access to intelligence. But if that investment doesn’t show up as better understanding, better decisions, and better outcomes, then it’s just extractive. We’ve already seen how that plays out.

If we don’t deliver medical-grade tools for patients deliberately and responsibly, they’ll build their own workflows out of whatever is available. In an environment this saturated with misinformation, that’s a risk we should not be willing to take.

This is a narrow window. If we lock this down now, we’ll spend the next decade trying to unwind it. If we design for sharing and explanation from the start, we might finally align incentives around outcomes instead of access.

Appreciate you pushing on this. We need more people in product willing to say it plainly.

Jacob Madsen's avatar

Frankly, patients are overwhelmed because interoperability was built for billing not educating or enabling care. Patients get formats that check a box for compliance, but patient portals do more to expose the shortcomings of EHRs than they do empower patients or their care teams.

Portland Paul's avatar

• “Agree on the cognition layer. My concern is failure mode. A polished narrative with nice UI can make an incorrect inference feel more trustworthy than raw data.”

• “Patient copilots should default to ‘questions to ask your clinician’ and ‘when to seek care now.’ Avoid differential diagnosis or med changes.”

• “Every summary should show provenance. What sources were used, what is missing, and confidence bounds in plain language.”

• “Hard stops for high-risk content: chest pain, neuro deficits, pregnancy, pediatrics, anticoagulants, self-harm. Route to human escalation.”

• “Guardrails must be testable: pre-deployment evals, post-deployment monitoring, drift detection, incident response, rollback.”

Eugene Vestel's avatar

great comment !

I largely agree. The reality is providers are structurally constrained time, reimbursement, liability, burnout. That isn’t a moral failing; it’s the system. If patients aren’t empowered to carry cognitive load, we’re guaranteeing inequity by design.

Where I’m aligned with you is that information asymmetry is the core problem. When only one side has access to synthesis, context, and interpretation, the other side is forced into passivity. And in practice, patients and caregivers often do the deeper work: tracking symptoms, noticing patterns, translating guidance into daily decisions. They live with the condition; providers visit it.

AI absolutely can level that field but only if we’re honest about failure modes.

manuel albarracin's avatar

Brilliant post. Couldn’t agree more.