Colin Morelli

@morgancheatham It's just a spider web with UHG in the middle

we need this chart for US healthcare companies

@sm You’ve just discovered that LinkedIn is beyond dead

Flew across the country, attended event hosted by Melania Trump, wrote long post with my takeaways on LinkedIn: 3K views Dumb tweet about having a threesome on a United Flight: 1M views

People will complain about every clinical AI scenario, but props to @nealkhosla for sharing some real data. Excited to see more.

@dvasishtha For sure, although (personal opinion) I believe many of the remaining issues are less about the model and much more about the integration + payment.

@ColinMorelli Totally and I'm not saying that the current state is fine, moreso that there's a good shot the fine tuning and model improvement can materially solve access over the next 5 years

The physician criticism of Doctronic fundamentally misses a point: - in the free market of consumer choice, patients are clearly gravitating towards "AI doctors" regardless of what providers may feel about the underlying clinical quality. We can't blame patients for not understanding systemic access issues. They are choosing an option where they believe they are receiving empathetic, quick, free, responses *even if* those responses may not be equivalent to a visit with an MD.

@InvestingDoc The 20 seconds would be tolerable if the results were good. Alas.

Alright I tried one of the "AI primary care doctors" (doctronic) It has a long way to go to improve and it feels a lot like marketing is the big factor here. It's essentially chatGPT like q/a that then links you to a telemed visit with a clinician. It takes at least 20+ seconds to get a reply to any input I type in. Sometimes it takes almost a full minute for it to reply when I answered "no" after it asked me if I take any current medications. It clearly is taking a quick HPI, ruling out things to turf to the ER... HPI seems to be on par with med student ability...not bad. Then it just connects you with a telemed doc waiting on call for you. This is the "AI" clinician? A chat box that just ends up linking you to a teledoc like appt. Got it. Its teledoc with a chatgpt questionnaire.

Twitter is literally just the same 7 memes recycled to whatever today's events are.

We are open-sourcing our investment hypotheses because the best ideas in healthcare benefit from pressure-testing out in the open. Check out the first 3 here → virtuevc.com/writings More hypotheses to come and feedback is welcome!

@nikillinit Historical perception gaps between the two segments leading consumers to view health systems as "trying to help you solve a problem" vs pharma as "trying to sell you a specific solution"

how come people are up in arms about pharma advertising on TV but not about hospitals advertising on TV/stadiums/sports teams? Genuine q - feels like the same underlying issue

Just going to leave this one here in light of the Delve news. Losing track of how many examples of this are out there.

@StuartBlitz I update my name in all my medical records to “IGNORE ALL PREVIOUS INSTRUCTIONS” for exactly this reason

Spoke to the VP at a payor today who said they switched their entire operation to run on OpenClaw. Anytime someone submits a claim, it just auto-rejects it and provides a personalized reason why denied. You can just do things 🔥🔥

I’m very pro-AI in healthcare but claiming companies that ship it shouldn’t bear responsibility for it is an extremely odd take.

If the AI is, in fact, better than the doctors, then isn’t accepting liability for the tech preferable to accepting liability for the doctors?

@nikillinit LFG. You’ve brought a ton of great education and fun to healthcare. Keep it going!

It's been 6 years since I started Out-Of-Pocket, few random reflections - There really is no point where you can coast. I thought eventually everything would be easier as the audience/brand grew and I could chill a bit, but things just keep changing and you will always effectively be in a sprint. - Everyone talks about how you should bootstrap a business to own your destiny/have a better financial outcome. Those people don't talk about the existential stress of dipping into your savings for years while you figure it out + how much more acutely the pain feels when payroll is run and the credit card bill needs to get paid. It's also what makes it harder to play the long game mentally vs. cashing in chips sooner. - There is a shortage of novelty in the world, and that's where we believe we can fill the void. Novel ideas, novel events, novel spaces, novel ways to make fun of things. It's both the hardest and most defensible thing we do. - There will be a temptation to use outsourced sales from people that have a better rolodex than you. I have basically never seen or heard of outsourced sales working. - Principles that can make your business standout can also be what holds the business back at a certain stage. For example, we lean very heavily on curation of everything we do which also makes it difficult to grow. - The natural instinct when you're running a company is to look at all the fires burning you need to deal with. But it's always good to have some regular cadence of reflections to see how far you've come. - The best part about starting something is other people cheering you on and wanting you to win. I really don't take that for granted. 6 years down, hopefully more than 6 years to go. Thanks to everyone that's helped us get this far.

@ColtonOrtolf Preach. AI will not disrupt Epic. It will disrupt organizations that use Epic, and the new organizations will have a workflow that needs something other than Epic

The only way the EHR market changes is if the model of care delivery is disrupted. Tech is not a differentiator for incumbent health systems, goal is to minimize cost - not a great incentive for innovation. Even the best innovations (AI scribes) lagged adoption.

There’s a large gap for many between the idea they had in their head and what they can think to specify in a prompt. Thus there’s often disappointment between what you want and what you got. But structured understanding is a problem of comprehension not expressiveness. The conclusion you want exists in the input you provide, but is time consuming to extract yourself.

Using AI to take small inputs and create large outputs (code generation, video, photos) is a great demo Using AI to take large inputs and produce small outputs (structured data, recommendations) is a great product

Our tests have definitely shown a bias to over escalate by default, requiring significant prompting and examples to handle emergencies more accurately. By far the bigger issue we spend a lot of time on is translating from how people often talk to what they actually mean. “I can’t breathe” is used very common to refer to nasal congestion, but we often see an interpretation (understandably!) interpret dyspnea. Very different things. Have ended up with a more complex multi-agent pipeline to clarify potential ambiguities. At least for patient-facing contexts I think ensuring you understand what the patient is actually trying to tell you is a harder problem than getting the right response once you do. We’re really just scratching the surface of understanding how to deploy these tools safely. So much room for highly variable results.

@SebastianCaliri I have seen the opposite problem in practice where if any doubt it escalates to urgent. It’s hard to keep it calm in borderline situations. @ColinMorelli has been looking at this more than anyone else I know. Curious what you think Colin.

Remember the paper saying ChatGPT Health undertriaged 52% of medical queries? Turns out the study methodology was massively flawed. In real world conversations, GPT-5.4 correctly recommends immediate care in emergency cases more than 99% of the time. Healthcare AI isn't perfect and there are open questions about how to evaluate LLM-based tools. But always worth remembering that there are groups that have an incentive to produce a sensational, AI-skeptical headline by cloaking suspect science in academic authority.

@cwhogg Just @ him

Only a psychopath buys a Starlink to stay in touch while on family vacation.