OneOncology

This is why we need 340B reform and accountability.

Andie Schilstra is #hiring . Know anyone who might be interested?

The Caregiver Is Often the Real Customer in Healthcare AI Every consumer healthcare AI deck I’ve seen in the last six months centers the patient. The patient asks questions. The patient gets summaries. The patient tracks symptoms. The pricing models, the engagement metrics, the user research — all aimed at the person with the diagnosis. This misreads the actual buying decision in chronic disease. The daughter managing her mother’s chemo calendar is the engaged user. The spouse weighing the CHF patient every morning and titrating diuretics is the engaged user. The adult child coordinating three specialists, two pharmacies, and a Medicare Advantage plan for a parent with dementia is the engaged user. They are also disproportionately the ones with disposable income, smartphone fluency, and a willingness to pay for relief. Patients in active treatment are exhausted. Caregivers are activated. The numbers tell the story. AARP estimates 48 million unpaid family caregivers in the US providing care worth roughly $600 billion annually. The average caregiver spends 26% of personal income on care-related expenses. Sixty-one percent are women. Most are juggling caregiving with full-time work. They are searching, scheduling, advocating, and decision-making on behalf of someone whose cognitive bandwidth is consumed by illness. This audience has specific, unmet needs that current consumer healthcare AI does not address: • Multi-patient context. One account, mom and dad’s records, separate care plans, shared escalation logic. • Asynchronous communication. The caregiver needs to ask at 11pm what the oncologist said at 2pm. • Decision support framed for non-clinicians. Not “your creatinine is elevated” but “this lab result usually means your father needs more fluids; here’s when to call the doctor.” • Care coordination artifacts. Visit summaries the caregiver can forward to a sibling, a home health aide, or the next specialist. • Permission and identity layers. HIPAA-aware proxy access, not a shared password. Consumer health AI built around the patient defaults to a single-user, single-condition, English-fluent, digitally-native, in-treatment model. That is not the chronic disease market. That is a sliver of it. The caregiver economy is the larger willingness-to-pay segment. It is also the segment most likely to convert engagement into retention, because the work does not end when the patient feels better. It ends when the patient dies. Build for the daughter. The patient gets served as a consequence. #HealthcareAI #Caregivers #ChronicDisease #DigitalHealth #FamilyCaregiving

Five things I’ve built with AI lately — some at work, some personal Yes, I use it to write emails, analyze documents, and answer questions I’d otherwise have to Google for an hour. But the most fun I’ve had with AI lately is actually building things with it. Here’s what my team and I have put together recently: 1. Automated a manual data entry process at work Simple tool, self-contained — but it eliminated a genuinely tedious step for real people on our team working with regulatory documents. Sometimes that’s enough. 2. A data migration tool between two platforms we use every day Yes, a standard import tool exists. But this one pulls and maps the data for me and flags when values don’t translate correctly. Runs locally, works well, and saves my eyes and my sanity on large detailed datasets. 3. A training and access prototype Always deprioritized because the ROI was hard to justify. AI made it cheap enough to stop waiting and just try. Pretty confident it’s going to be a big win for our sites. 4. A Claude branding skill for Hometeam Pizza A reusable AI setup that knows our voice and vibe. Creating content and generating ideas is so much easier with this as a starting point. 5. A recipe sharing app for my family Meal planning, AI-generated recipes, a comment thread for swapping tweaks. A fun project I’m learning from that also helps me personally. That’s enough for me. I don’t think AI is replacing software teams. But I do think it’s going to compress timelines, help people actually communicate what they need, and — maybe most importantly — finally give teams a way to build the tools that are always important but never urgent enough to prioritize. That last one is where I think the real opportunity is. The backlog of good ideas that never got resourced is about to get a lot shorter.

It's not perfect but I've been building a slick workflow for my agents. Human in the loop is important but how do you condense the amount of time a human needs to be in the loop and has the highest leverage in that loop? I call it Human Approving the loop. Or Human ON the loop. Not in it. The loop runs without me. My primary goals have been: 1. Reduce cognitive functions needed to review massive amounts of inbound data. 2. Remove decision processing and routing of incoming new information. 3. Producing artifacts from new information. 4. Taking action on those artifacts. 5. A reinforcement loop that get's better over time. I own all of this 100%. If Obsidian, Claude, Codex blow up tomorrow, I own all the knowledge, data, structure, workflows, etc., and I can use open source models. Literally any model can pitch in to build my wiki.