I’m blessed to go to a lot of healthcare IT conferences. I’ve heard just about every keynote speaker imaginable. I’ve heard amazing keynote speakers. I’ve heard inspiring keynote speakers. I’ve seen entertaining keynote speakers. I’ve seen stars on stage. I’ve seen political figures on stage. I’ve seen terrible speakers. What’s interesting when you hear a keynote speech from someone great is that you generally don’t want to hear them again. Not to tell you how the sausage is made, but they often do the same presentation with very minor tweaks for the specific conference.
What’s interesting is that one of the most interesting and thought provoking keynotes I’ve seen was Zack Kass at the MEDITECH Live conference last year. When I saw that Zack Kass was keynoting HFMA’s annual conference just months later, you’d think I’d be running for the evening social events early. However, it was just the opposite. I couldn’t wait to hear Zack talk again. For those not familiar with him, he’s a former executive at OpenAI (ie. ChatGPT) and currently researches, teaches, and publishes on generative AI and related trends. I was excited to hear him again because I’m confident that I didn’t process everything the first time. Plus, AI is changing so quickly I wanted to see if his presentation evolved just as quickly. Not to mention, the Q&A is always different and insightful.
The good news is that his presentation had evolved. Some of the “this may happens” had become “this is happening.” Plus, he frames what’s happening so well. However, one thing he said really stuck with me and keeps rolling around in my head.
A new internet is being built for agents.
He pointed out a number of things that illustrate this point. The most poignant to me is that agents don’t care about the UI the way humans do. Much of what happens on the internet is to make it friendly and useful for humans. An agent visiting a site doesn’t care about the UI.
The easiest example of this is an ecommerce site like Amazon. If I sent an agent to go to Amazon and make recommendations for a new microphone I need to buy for my podcast, all that the agent needs is data. That’s probably a bunch of XML files as opposed to a rendered web page. In fact, the rendered web page slows down the agent.
This illustrates why an agent needs an entirely different internet than what we have today. Ok, maybe it’s a little much to say internet since it will likely use our current internet to do all of those things. However, does every website need an agent friendly site the way we worked so hard to create mobile friendly websites? Another bad comparison since mobile friendly websites were still catering to humans which have the same needs just on a smaller screen. Agents and bots that visit your site need something quite different.
What does this mean for healthcare?
It’s easy to keep extrapolating what this new internet looks like and provides us as consumers, but what does it mean for healthcare. Sadly, I think this is probably an example of where healthcare is going to fall behind other industries. I find this interesting since in some ways healthcare is pushing the use of AI with things like AI medical scribes and even robotic surgery in line or ahead of other industries. I’m sad to say that I think healthcare is going to be really slow to embrace this new AI agent centric internet.
Let’s think for a minute about what that looks like. An AI agent visiting a hospital and health systems website doesn’t need the UI either. It just needs the list of doctors, what services they provide, their ratings, etc etc etc. The agent needs an interface where it can query for eligibility of a patient to be able to be seen by that healthcare organization. A healthcare agent would want to be able to schedule an appointment based upon the available appointments. An AI agent would want to easily access all the data from the patients full patient record including things like lab results and prescriptions.
I could go on, but I think you get the idea. All of the above items that an agent could find useful to be able to perform some action for a patient would be amazing on this new AI agent focused internet. However, think about how many of the companies that control this information are going to want to embrace an AI agent engaging with them. You can see inklings of this idea with things like Smart on FHIR that try to start building the structure for this AI agent internet. However, that definitely falls miserably short from what’s needed to make the above AI agent internet a reality. That’s true for all of the healthcare standards and APIs.
Plus, I think most of these companies that hold the data are going to hide behind privacy rather than building this new form of accessibility.
The crazy thing to consider is that the technology is moving so quickly that startup companies aren’t going to wait for this new AI agent internet to be built. Agents have gotten quite good at navigating websites, scraping information, and basically acting like a user. Entire companies have been built on the back of screen scraping technology because the systems of records [arguably?] haven’t made the data available in another way that was financially viable for them to access it. Follow Brendan Keeler on LinkedIn to see some of the many cases he’s tracking that are dealing with these issues.
We’ll see how the court cases play out, but when you add in things like information blocking it’s hard for me to imagine a future that doesn’t allow startup companies access to the health data and have AI agents engage on behalf of the patient. These companies scraping your data and engaging with your company through your web UI has to be one of the most inefficient engagement mechanisms that exist. The bots can scale so quickly that they can basically create a denial of service attack on your web UI doing legitimate requests. However, I think this is what will happen given no other options.
Faced with this reality, what will healthcare IT organizations do? I feel like they’ll have no choice but to build the AI driven internet that I described previously.
The good news is that this isn’t happening tomorrow. However, it’s happening quicker than most of us want to admit.
Think about it from my kids perspective. Do you think they’re going to wait on hold for a doctor’s office? Definitely not. They’ll be asking their bot to make the phone call, text message, portal login, etc to do it all for them. That’s right. They aren’t going to wait for healthcare to build the infrastructure for them. AI is moving so quickly that the AI will navigate whatever archaic technology you want to use including a phone call.
Does that mean that you don’t need to worry about changing? I guess in some cases that may be fine for a while until the bot starts realizing which organizations are easier to interact with and prioritizes your patients to ones that have more modern options.
What do you think of the idea of a new Internet for AI agents? How will this play out in healthcare? Let us know on social media. As you can see, I’m still trying to figure it out myself. However, Zach Kass definitely caught my attention.
