What I’m Doing Now

Last updated: Tuesday, February 10, 2026

This is a now page. It changes when I do.

Quick Snapshot

I’m building tools and communities that help patients use AI to understand their own health data, on their own terms.

SMART on FHIR + AI: Turning Health Records Into Conversations

The most exciting thing I’ve done recently is something that shouldn’t have been this hard. My dad is 95 and has been a patient at Stanford for 15 years. That’s 326 encounters, nearly 1,800 observations, and 247 medication records. We have a legal right to all of it, and thanks to open standards like FHIR, I can download it. But the file is 111 megabytes of JSON. Every AI tool I tried choked on it.

Josh Mandel, who helped build SMART on FHIR, created something called Health Skillz. It packages your health records as a FHIR bundle you can download locally. The key insight was simple: instead of uploading the data to the cloud, bring the AI to the data. I used Anthropic’s Cowork with Opus 4.6, to create a powerful, local desktop agent, and pointed it at the folder on my Mac. No upload needed. Claude read the files directly from my hard drive.

One prompt. That’s all it took. Fifteen years of records, organized, summarized, and queryable. I can ask about medication changes over time, track lab values across years, and show up to the next appointment prepared with real questions based on real data.

This is what patient empowerment looks like in 2026. We fought for years to get patients the right to access their health data. We got FHIR, we got open APIs, we got federal rules that say providers have to share. But access without comprehension isn't real access. What I did was turn a 111-megabyte wall of JSON into a conversation. The tools exist. Right now.

AI Patients Community

This is probably the most important initiative I have going. AI Patients is now live. It’s the start of a community of patients and caregivers learning to use AI as a thinking partner for health, not to replace doctors, but to ask better questions, spot patterns we’d miss, and show up prepared.

As I write this, we haven’t met yet. But we hope to meet periodically online and perhaps in person in the Spring and Summer. People share short talks about how they’re using AI for health, what they tried, how it went, what they learned. No polished presentations. Just honest conversation among people figuring this out together. If you’re curious, sign up at aipatients.org.

Operationalizing Critical AI Health Literacy

Both of our 2025 papers are now published: Critical AI Health Literacy as Liberation Technology in NAM Perspectives and Generative AI as Third Agent: Large Language Models and the Transformation of the Clinician-Patient Relationship in the Journal of Participatory Medicine. The Critical AI Health Literacy paper was very well received and got a lot of attention.

My main intellectual focus for 2026 is operationalizing the framework. I want to move from theory to implementation, helping patients actually use AI in ways that work outside institutional constraints. If you’re an organization, funder, or collaborator interested in supporting this work, please reach out. This is where I’m putting my energy.

Tools I’m Using

I’ve moved past the “experimenting with everything” phase into a settled workflow. For anything involving my local files and health data, I use Claude for Chrome and Claude Cowork. For research and information tasks, I use mostly Gemini, Notebook LM, and ChatGPT. For voice input across all of it, I use Wispr Flow, which lets me dictate into any app, cleans up my speech in real time, and handles both English and Portuguese. The tools that have actually stuck in my daily work are Claude’s local tools.

Ideas I’m Working With

  • The gap between data access and data comprehension. We won the fight for patient data access. FHIR exists. Open APIs exist. Federal rules exist. But a 111 MB JSON file is not “access” in any meaningful sense. The next frontier is comprehension, and AI is the bridge. I’m documenting what that looks like in practice.

  • Community as infrastructure. Individual patients using AI is powerful. A community of patients sharing what works, what doesn’t, and how to think with AI is something else entirely. That’s what AI Patients is becoming.

  • Local-first AI. The most powerful thing about Cowork isn’t the model. It’s that the data never leaves your machine. For health data, that matters enormously.

If any of this resonates, I’d love to hear from you.

Let's Work Together