Healthcare Needs Its Cursor Moment
Innovation always hits a wall when technology isn't easy to use
Cursor doesn’t feel like it should be a billion dollar business. Case in point: most of the readers here have probably never heard of it. It’s an app that copied another app (VSCode, the Microsoft Word of coding), and slapped a ChatGPT window on the side. They’ve developed a bunch of features that make it easier to work side by side with everyone’s new best friend, and have built their way up to a business valued at 2.5 billion dollars.
I write this with no disrespect intended to Cursor. As someone who grew up seeing the release of iPhone, YouTube, ChatGPT, and more, I wholeheartedly believe it is one of the most impactful pieces of technology I have ever seen. It’s helped me, someone whose only coding background was a day in college where my friends and I decided to impersonate each other in our respective classes, build and test minimum versions of 5 different ideas I’ve had over the past few months.
What does Cursor get so right? The answer is surprisingly simple. Having AI, or any technology for that matter, deeply integrated into your environment, workflow, or digital ecosystem makes a world of difference. The rote process of going to ChatGPT, pasting in your code, asking your question, copying its response, and then going back to your editor and pasting it in gets really old, really fast. Especially after you’ve used a software where you can just highlight the error, hit the edit button, and sit back and relax while AI puts on a show.
This ease of use overshadows everything - even quality deficiencies. I personally think v0.dev is way better than the models Cursor relies on, but the process of downloading and copying stuff over from v0 into my local editor starts to become such a slog that often I find myself relying on Cursor even when I know it’s not the sharpest tool in the shed. That’s a moat, if I’ve ever seen felt one.
What does this have to do with medicine? Well, over the past 20 years there’s been an explosion of point-based solutions for clinical problems, billing workflows, and everything else that’s behind the scenes of your yearly doctor’s appointment. Most will suffer from the same sword: it’s just too inconvenient for healthcare workers to spend their days shifting back and forth between 10 different apps that each solve a unique problem.
One of my responsibilities at the cardiology clinic was helping to manage the remote monitoring of patients with implanted devices (pacemakers, defibrillators, rhythm recorders, etc.). This involved logging into 5 different portals on a daily basis and individually downloading and printing transmissions from each one. This process was so frustrating that once, when a rather prominent startup came to pitch us their remote blood pressure monitoring technology, we had a rather comedic moment where we all agreed on its clinical utility but said “Sorry, no thank you” just because it was one more login to remember. I remember pitching a simple idea to my CS friends: a universal platform that brought all the data from these various sources into one dashboard. No new technology or added features or anything of the sort - the value would solely be that folks now only need to login to one platform. Unfortunately, in a classic scenario that’s always simultaneously frustrating and validating, multiple successful companies had already beat us to this.
Cursor reminded me of these insights, and reaffirmed my conviction that healthcare will benefit from unified, integrated platforms that enable seamless interoperability between our growing lists of technologies and tasks. The domain which seems to have wholly adopted this already is AI assisted image interpretation, with things like PathChat for pathology, EchoNet for cardiology, and what feels like every single med-tech company in the world for radiology. These have all prioritized directly plugging into existing image acquisition or study reading technology, which I predict will vastly improve their uptake amongst physicians and will significantly decrease the burden of using them.
For things like insurance and billing, practice management, and even medicolegal work, there’s a ripe opportunity for an AI-first ecosystem that makes it easy to call on different tools. It seems like I hear about a new automated prior authorization or revenue cycle management company every other week, but I’ve yet to see many that natively fit into existing softwares or easily transfer data over from other sources. Maybe there’s a selling pickaxes during the gold rush moment here: an iPhone with an App Store type of thing for the admin side of healthcare, where it’s easy to download solutions and plug them into your existing infrastructure, but also easy to delete them and quickly try out new ones.
Clinical decision support is probably the field that can take the biggest lesson from Cursor. There’s a bunch of tools out there that help doctors find medical literature, draft differential diagnoses, and even get direct answers on how to treat patients. However, none of these integrate well with the electronic health record, which is the doctor’s version of VSCode. Here’s why that should change (and how Cursor changed it for coding):
It’s incredibly inconvenient to send context to the tools. No one wants to export or copy/paste medical records from the EHR into a medical record summarizer - at that point, you’d rather save the money and spend the same effort on summarizing it yourself. (Add to Chat Button).
It enables functionality even when the tool isn’t explicitly called. It’d be so cool if AI popped up on the right side of a patient’s chart with a small note that said “Based on reviewing this entire chart, X trial seems particularly relevant to this patient’s care.” (Autocomplete)
Doctors already spend 6 hours in the EHR for every 8 hours of patient care, and those 6 hours are also the main times that they’re running into the use cases for all these tools. You’re most suited for help drafting a note while you’re actually writing the note. (Forking VSCode)
I recently learned that the process of selling a product to a hospital often takes 12-24 months at minimum, just because of how difficult it is to implement new tools into healthcare workflows. Amidst all the jokes made about ChatGPT wrappers, the idea of taking existing technology and packaging it in a way that decreases this implementation burden certainly solves an unmet need. Cursor did this for coding, and it’d be valuable for someone to do it in healthcare.