The Information Exchange: Standards-based Thruple Edition

All the kids went back to school this week, which means so did we! Fall is the busy season in this industry, with a long syllabus already lined up. \* Brad went to DC for the one year anniversary of the CMS Health Tech Ecosystem and came back with a read on where it stands. \* CMS-0057 goes live Jan 1! Pryce (who has spent a lot of time in the prior auth weeds lately) is nervous. The million subsystems in a trench coat that are payers has many groups in a spec-off. We also cover the fall reading list (HTI-5 this month, HTI-6’s ambiguous fall launch, and more), why enrollment is the unsolved problem that keeps apps off FHIR, and Brad’s mid-pod theory about blood pressure cuffs shipping with cellular radios and a TEFCA connection. All that and more in a fun one. Let’s dig in. Relevant Articles \* CMS’s Health Tech Bombshell**:** The original read on the Health Tech Ecosystem pledge, now one year old and the subject of Brad’s DC recap. \* Artificial Scarcity, Meet Artificial Intelligence**:** Prior art about MCG and Interqual \* The Regulatory Roadmap Reappears: The Unified Agenda drop we mention \* HTI-5: When the Scorpion Learns to Swim: The deregulatory rule finalizing this month, including the information blocking provisions EHRs aren’t psyched about. \* A Shot Across the ONC’s Bow**:** Will the EHRs clap back about HTI-5 information blocking provisions? We think yes. \* An Argonaut Miss?: Title doesn’t state it well, but ONC needs to think broader than just the Argonaut Subscriptions IG. Solve the enrollment problem or you’ll have another five years of apps complaining. \* CMS Launches ACCESS**:** Had to dig for this one, but this is background for Brad’s on-the-fly theory about devices, GLP-1 monitoring, and where the payment models point. Chapters \* **Intro (0:00 – 0:47):** We’re back and so is the school year, regulation, and a splash of litigation. \* **A Year of the Health Tech Ecosystem (0:47 – 5:19):** Brad’s DC report of the CMS Health Tech Ecosystem anniversary event. \* **Four Months to January First (5:19 – 11:13):** We go deep on CMS-0057, including the hidden complexities that are causing so many organizations pain. \* **The Light DTR Casualty (11:13 – 13:52):** HTI-4, buried in the inpatient rule, killed the SMART app path and yoked prior auth to the EHR. \* **Could 0057 Have Been Better? (13:52 – 17:52):** Pryce weighs in on whether there was a better path for the prior authorization roll-out. \* **The Enrollment Problem (17:52 – 21:18):** The fall reading list, including why Argonaut’s patient subscription still leaves 99% of apps unable to be FHIR native. \* **Blood Pressure Cuffs with Cellular (21:18 – 23:20):** Brad’s mid-pod theory on BALANCE, TEMPO, ACCESS, ELEVATE, and devices getting roped into certified health IT. \* **FMK: Subscriptions, Bulk FHIR, Agents (23:20 – End):** We go around the horn to determine the fate and future of American interoperability with some hard decisions. Transcript *We ran the transcript through an LLM to smooth it out. So it’s a rough approximation of the conversation (and in many cases significantly clearer than our rambling), but notably may diverge from the word-by-word blows quite a bit.* **Brendan Keeler (00:00):** All right. In the words of Billy Madison, back to school, back to school, to show the listeners we are not fools. It’s school time. Ryan, your kids are going to kindergarten? **Ryan Tucker (00:11):** Pre-K two, pre-K three. Very serious stuff. **Brendan Keeler (00:13):** Pryce, your kids pretty soon. You have a broader array of ages. **Pryce Ancona (00:18):** They started yesterday. We’ve got a third grader, a first grader, and then a two year old at the pool right now just rocking the goggles. He’s chillin’. **Brendan Keeler (00:28):** Well, more broadly, we’re all getting some homework, because we’re headed into a busy season of regulation, of federal litigation perhaps, of lots of stuff coming to close out the year as we head toward the deadlines. The test comes January first for many payers. But before we get into all this, Brad, you went to DC to meet the teachers, did you not? **Brad Thorson (00:47):** I did. We had the one year anniversary for the CMS Health Tech Ecosystem. This is the fourth event they’ve held in Washington so far, fourth or fifth. A couple things really stick out. One of them is that Amy Gleason has hit her stride. She’s not in this to be Usain Bolt. She’s going to be the marathon runner who breaks the world record. They’ve now got over forty people working in product, they’ve continued to expand the number of categories they want entities to participate in, and one of the most noticeable things at this event versus the others is that the big legacy players showed up in force and they are pledging to more categories than they maybe initially wanted to. That’s a good sign that the industry is adopting the idea that coopetition is positive. One of the really big takeaways for me is that there are polarizing figures in this administration, and Chris Klomp was absolutely perfect in articulating the vision of the health tech ecosystem. He identified that this has nothing to do with politics and is really about returning control to patients. I thought he was the best speaker by far at the entire event. It’s awesome that he’s moving into higher positions of power, because I’d watch that guy talk about health tech all day long. No offense to anybody else who speaks about it. I just think he’s fantastic. We’re seeing all of the major EHRs show up and figure out how they want to participate. We’re seeing some of the newer kids on the block, who maybe didn’t have as big of a health tech footprint beforehand, getting deprioritized or seen as necessary but not sufficient actors in the space. That first category of tools depended on patient identity, so CLEAR and ID.me were omnipresent, but we saw a lot more of your Epic, your MEDITECH, and so on. And we got payers. Most of the major payers were present, which was again very cool to see. I’ve got to give a shout out to b.well for committing to every single category, just so that Liz Lewis can’t sleep at night and has to run every work group. It has come a super long way. We’re seeing a lot of maturity. The big question is going to be breadth versus depth and whether they can do both. Amy seems confident they can, and the next year will really be about which of these categories stick and which ones they continue to invest in. **Brendan Keeler (03:16):** It’s such a dichotomy across the board. Tremendous successes and things that aren’t real, all mixed together. You have them threatening regulation on one hand, like you’d better go do this voluntarily or we’re going to make you do it three times over. And then you have Oz on stage asking why aren’t you more like Uber, why aren’t you going and breaking laws. There are all these fun ironies, but they’re getting s\*\*t done. They’re going to put some of these capabilities in HTI-6 to make it happen. Industry has proven out many of them, so then they get to say, okay, this gets the pilots out there, and now we’re going to get everyone to do it. It seems very logical that that’s what we’ll see. That’s what struck me watching the recording, the bipolar nature of it. **Brad Thorson (04:07):** The other thing I felt seeing big names present in that room takes us way back. I started a health tech company almost a decade ago and I knew basically zero about healthcare. It was really, really hard for me to find people who would teach me how things actually worked, so it was a process of trial and error. An undernoted and very cool thing about the Health Tech Ecosystem is that if you’re willing to participate, no matter the size of your entity, you can be on calls with executives from Epic or MEDITECH and understand their workflows and requirements. I don’t know that we’re going to be able to measure this, but if we could, I think we’ll see an acceleration of startups understanding traditional healthcare workflows for things like prior authorization. Their previous approach may have been, I’m just going to screen scrape everything and break the system. Now they understand why the system works the way it does and can start to operate inside of it. **Brendan Keeler (05:10):** So the teachers made a good curriculum for us. Is that what I’m hearing? **Brad Thorson (05:14):** Yeah. We’re going back to school and we’ve got a great syllabus. **Brendan Keeler (05:19):** One of those pieces of the curriculum is prior authorization. We already have the CMS-0057 deadline on January first. There’s 0062 hanging out as a proposed rule, and now the CMS Health Tech Ecosystem has a prior authorization work group. Pryce, you’ve probably been the deepest of anyone on the team working with payers and delegated vendors on prior auth. How are we going to do? **Pryce Ancona (05:44):** I’m nervous. Everyone working in the 0057 space right now is feeling like holy cow, I can’t believe folks are going to try to start using this in production on January first. It is an administrative workflow, so it’s not a patient safety escalation to me. I’m not going to start shouting from the rooftop that everyone needs to protest outside the Humphrey Building and get 0057 pushed back a year. The biggest thing I think people don’t realize is that we act like the prior authorization problem is health systems needing to speak with payers. Right now it’s a very administratively burdensome process, and it’s a terrible time in the life of the patient. If they’re getting permission from their insurance company to do something from a medical perspective, they might be sick or anxious to figure out why they’re sick, and it takes too long. But we simplify it to the EHR should be talking to the payer systems. Three easy steps, and doctors love these three easy steps: you figure out if you need prior auth, you figure out what information you should be providing your payer so they can determine authorization, and then you submit the actual case. That’s framed in a way everyone can understand. What people don’t understand is that each of those EHRs then has to figure out to whom am I connecting. Is it my top five payers, or am I going to connect to my clearinghouse and my clearinghouse is going to promise to facilitate the routing of this request to the right payer? Then it gets to the payer, and it’s a clearinghouse who maybe has never dealt with prior authorization exchange before, or never using FHIR and the DaVinci standard. They’re a layer that receives it and perhaps passes it to another system which does CRD, which might talk to another system which houses the questionnaires for DTR, which then might pass the case to a third system that’s one of ten utilization management delegated vendors at that payer. It’s one of the first instances in a very long time in health tech where we’re saying everyone should start using this functionality on this date, and there’s a gajillion permutations of how and to whom you would connect. Any slight variability in how people interpret the specs, or let’s say one clearinghouse has a bug, and it’s going to be millions of prior authorizations that fail electronic transmission. And then what? So I’m concerned from an operational perspective. The tech will work itself out. What have you all been hearing from folks who are really in the prior auth weeds right now? **Brendan Keeler (08:31):** I love Scott Rossignol’s idea, which he posted somewhere, or maybe it was Josh Mandel. Most people gravitate to this. Wait a minute, if we just had payers post all of their policy publicly, we could run an LLM over it and just have it. Instead DTR is here’s a little bite-sized chunk, oh, you want more, here’s a taste, here’s a little taste. That back and forth has a million rough edges that can break. Even in the beautiful scenario where you take away the fact that the payer system is a million subsystems in a trench coat, pretend it’s just one system. Looking at the back and forth of client and server and server and client and subscriptions and CQL, it’s going to break. There are a hundred edges where it will break, and it’s just going to suck. The idea of front loading it so you can shoot over the documentation you need, or making it a simpler transaction, makes a lot of sense. We’re not going to get that, but if you went back in time and reinvented this, that’s what you would do, MCG and InterQual be damned. That to me is logical. We don’t exist in that world. We exist in a world where there’s a million subsystems that haven’t really been accounted for and need to play in this space and meet the deadline. **Pryce Ancona (09:46):** Just to be clear, some of the biggest players in the industry are saying, okay, we need to get this set up, let me see your specs. And whomever they’re talking to says, yeah, let me see your specs, let’s make sure our specs match. And neither side has specs and neither side has built it. Some payers are still figuring out who they’re going to use as the interop layer for their DaVinci implementation. This is supposed to turn on in what, four months. So it’s going to be a little heinous. The biggest concern for me is that what could have been a jump from terrible administrative burden to computers doing what they’re good at and making everything more efficient instead becomes providers having a bad experience for three months and then using it as little as they’re allowed to by their employers or by the MIPS payments they’re receiving or not receiving. **Brad Thorson (10:37):** Sometimes we flatten a workflow and its constraints and assume those constraints are directly related to motivations. Vectorizing a policy is creating an understanding of a snapshot in time. Payers have utilization management because they need to control spend. If there are easier ways to gamify that, they’re going to change the rules. This is a can of worms we maybe don’t have time to unpack today, but a major problem in assuming we can and should automate something is that we miss what the external friction is. **Brendan Keeler (11:13):** No, I think you’re right. Great, we made the transaction cheap, but the tension of are you going to approve this or not doesn’t change. It’s just the speed at which that decision gets made. The last thing that really grinds my gears about the situation we’ve put ourselves in is that HTI-4, buried in the IPPS inpatient rule last year, got rid of the light DTR option and really yoked this to the EHR. We’re in this era of we’ve got to empower alternative solutions and the EHRs are too powerful, and then we do regulation literally geared so that it must be in the EHR or in a supporting technology, where the EHR says okay, we’re using Darena as our chosen technology for prior authorizations. How can you go on stage and make these pronouncements and then get rid of the optionality for light DTR, which is more like launching a SMART app that does your prior authorization? Pryce, do you agree? You were working on this. **Pryce Ancona (12:10):** One hundred percent. To explain the paradigm for the audience: let’s say I’m a doctor and I place an order for an MRI. I should get an immediate response, if the back end systems are set up and connected, that says hey, you do need to request prior authorization for this order, and here are the questions the payer has in order to determine whether they’re going to authorize it. Those questions come across in the form of a FHIR questionnaire. At the moment, the EHR must support receiving those questions and showing them to the user, and the user answers them in the EHR. That’s called the native flow, where the provider stays native to the EHR. But there are vendors who do way cooler stuff with clinical data that helps you answer the questions way faster, and that could be a SMART app. Let’s say I place that same MRI order and boom, because my analysts set it up this way, instead of the questionnaire coming back into the EHR it automatically pops open a SMART app window with my questionnaire app. There’s a field in that one that says just drop all the documentation about Brendan and I’ll run AI through it and propose answers for you. I’m like, gosh damn, that’s so cool. That’s what generative AI should be doing. Somebody’s out there building it because all they care about is prior auth. Meanwhile the EHRs are getting the regulatory capture and aren’t forced to support that bolt-on system, which is the light DTR flow. So that’s the difference between native and light DTR. It’s a silly nuance and I guess just a bunch of nerds care about it, but in a year a bunch of providers are going to care about it. **Brendan Keeler (13:43):** We are those nerds. **Ryan Tucker (13:44):** Us. **Pryce Ancona (13:46):** So I’m bummed about that. But folks smarter than I wrote those rules, so we’ll see where they go. **Ryan Tucker (13:52):** I’m curious here, and we’re going to get into new regulation coming, HTI-5 and HTI-6, in more detail. Taking the lens of HTI-5 coming in, it was very deregulatory. It was trying to strip out the old news, the things that didn’t matter anymore. HTI-6 is now rebuilding that with more FHIR-based standards. But from what you’re saying about how 0057 is looking, we don’t know what’s going to happen once January first hits, and the signals aren’t great. It seems like the IGs might be a little too broad, or at least we don’t understand what’s going to happen behind the scenes once the message gets there. What could have been better for 0057? Was it stricter regulation? It seems like we’re not moving that way. Is it not building in the how, and instead focusing on the actual why behind the workflow? I don’t really know the answer, because maybe it’s you just need to meet this end goal and you can accomplish it how you want, but then that’s very broad and everybody will have a different interpretation and it’ll be hard to play together. Pryce, with that lens, do you think there’s anything that could have made 0057 better as we get to this point of what looks like it’ll be chaos? **Pryce Ancona (15:16):** Yeah, it would have been great if in the rule they mandated everyone use HTD consulting for the services they need. I’m just kidding. **Brendan Keeler (15:25):** We offer prior authorization services. **Pryce Ancona (15:30):** This next segment brought to you by. It’s hard to say, because the problem to be solved is part of the issue, Ryan, not the regulation or how they wrote it. Like Brendan said, take a bunch of different software systems who have never had to do challenging interop before. Maybe they’re all even small, because none of them are gigantic core administrative processing systems like TriZetto or gigantic EHRs like Epic. You’ve got all these green newbies going, I think we’re supposed to do something for CMS by January first. And then you gave them the hardest IG of all time to implement, hands down. So maybe it shouldn’t have been a boil the ocean approach. Maybe we should have started with CRD. Get your most health tech eager analyst to set up clinical decision support with regards to coverage requirement discovery. Then you submit, and you get to submit sixty percent of the prior auths you usually submit, because for forty percent of them you realize you don’t even need prior auth. Maybe what we needed was a little bit of stair stepping. But this whole administration, and this is what I was thinking while you were talking about the Health Tech Ecosystem, Brad, is throwing spaghetti at the wall to see what sticks. They’re going to see who succeeds and who survives, and then they’ll regulate, or there will be carrots and sticks they create in reaction to how the market adjusts. It’s a little bit ruthless. We just got thrown into the water by our belt loops and we’re going to figure it out. **Brendan Keeler (16:59):** All right, Tom Sawyer. And look, I appreciate the focus on the workflow. It’s just that the shape of it is contorted by these weird oddities we have in the industry, like MCG and InterQual not wanting to give up their policy because it’s their IP and their business. I don’t blame them for that. That is how they make money, but it contorts a lot of this into a less useful shape than if we’d thought about it de novo. What else are we thinking about for the rest of the curriculum this fall? What take home tests are coming? What books are we reading? **Pryce Ancona (17:36):** I don’t know, you tell us. You’re the TA. I’m forgetting. We have HTI-5 finalizing, and then Ryan was just saying we have a proposed HTI-6. When is the comment period over, or when does that finalize? **Ryan Tucker (17:51):** Is that toward November? **Brendan Keeler (17:52):** We know from the unified agenda that HTI-5 will be this month, that they’re trying to finalize in August. That’s the deregulatory rule where they deregulate away parts of information blocking in a way EHRs aren’t super psyched about. And then HTI-6, which we don’t know what’s in it. It’ll come out in November or something like that, but should have a lot of new API stuff, is what they’ve hinted. Federal lawsuit against Epic rumored to be coming. We don’t know what that’s about, but we shall see. And then of course the test on January first with prior authorization. Lots of fun stuff. What do you all want to see in HTI-6? **Ryan Tucker (18:30):** Lately I’ve been thinking a lot about agentic movement with AI. Brendan, what you brought up in terms of publish your spec out there and let an agent cater the transaction building and message building to what needs to happen, I think that would be really interesting. I know some of that was covered in HTI-5, there was some around agentic execution. But more FHIR-forward standards in general moving forward. That would be interesting to me. **Pryce Ancona (19:00):** FHIR subscriptions is what I’m most interested in, by a pretty long shot. But the subscriptions work group in the CMS Health Tech Ecosystem is just encounters and appointments. Encounters and appointments is a little bit too small for me to start jumping up and down for. Having an API you can subscribe to for proactive event notification is the most insane thing we haven’t regulated yet. **Brendan Keeler (19:27):** You make a really good point. There are many ways for ONC to F this up in HTI-6, so commentary will be important. One you just highlighted is whether there are the right triggers and data types. Subscriptions is really broad and can be used for a lot of different things. If they just have it for encounters, is it sending every new update, or just on certain triggers? Those details will matter in accomplishing a goal. I wrote an article a couple weeks ago about this. If they just look at Argonaut’s work and adopt the patient subscription, which is an IG that profiles subscriptions to accomplish a workflow, you could easily say all right, Argonaut never misses, let’s just copy that. They did great work focused around I am a patient app, or I have a single patient and I want to listen to updates to that patient. That’s supplementation. That’s not enrollment. What people yearn for is not better supplementation. Better ingestion of a patient once you know about them as relevant to you is great, hurrah. What we want is the ability to kick off a workflow, because we can’t poll right now. We can’t search across and say show me all new orders. We can’t listen and say show me all new orders, or let me listen for new encounters. Enrollment is the big unsolved problem that prevents ninety-nine percent of apps from being FHIR native. We need to make sure that’s in HTI-6, or we’re going to look up and go great, now we have slightly better supplementation and I still have to start with HL7v2 as my enrollment step. **Ryan Tucker (20:59):** So you’re calling for the death of ADTs? **Brendan Keeler (21:01):** Look, if we want to live in a world where it’s FHIR only, we’d need to go after enrollment. You can’t reliably assume an EHR has HL7v2 ADTs. From certification criteria alone, you don’t get a uniform enrollment step. There is no technology. **Brad Thorson (21:18):** Wait, but I have a prediction. **Brendan Keeler (21:19):** What’s that? **Brad Thorson (21:20):** What I’ve been doing while you’ve heard my keyboard typing: if you take a look at the health tech ecosystem, actually there are very clear motivations, things certain people in power want to see. We’ve also seen a ton of CMS programs launched, and it feels like it’s at a rapid pace. So I was trying to figure out what is the intersection of BALANCE, TEMPO, ACCESS, and ELEVATE. This group is smart enough to know they can use HTI-6 to unlock that. I’m mid research, mid-pod, so I’ll get you a better explanation after I can think about it more. **Brendan Keeler (21:57):** I’m like, man, is he writing us a SOW right now for a prior authorization deal for the payers and delegated vendors we help? **Brad Thorson (22:06):** I do think connectivity to TEFCA or these nationwide networks, as opposed to directly into the EHR, is it. There’s going to be something around network access and whether that ropes in HIEs or the new state based HINs, or whether it’s all at the national level. I would not be surprised if there was some effort for regulated devices to have connectivity, whether that’s cellular built in. There’s no way CMS is going to pay for GLP-1s and not try to track things like blood pressure and weight, which fits into the ACCESS model. My wild, out there hypothesis, based on three minutes of Googling and writing, is that we’re going to see devices that wouldn’t normally have fallen under certified health IT somehow get roped in positively. Give them an incentive: hey, if you add cellular connectivity to your blood pressure cuff, we know that’s going to get us better readings, and maybe it ships pre-integrated with TEFCA and we’ll increase your reimbursement by fifty cents a reading. HTI-6 and the payment models are not connected, but I think we’re going to see something. **Brendan Keeler (23:20):** We like those intersects. That’s good. All right, FMK. We’ve got FHIR subscriptions, bulk FHIR, and agentic access. Which are you feeling for HTI-6 in regard to those three technologies? Which ones would you keep or cut? **Pryce Ancona (23:43):** Agentic access in what form here? Are we saying it’s information blocking for EHRs to not allow agentic access, which is already true, or are you saying something else? **Brendan Keeler (23:54):** We don’t know. They’ve alluded to MCP or COIN as the technology some people have started to play with in the healthcare setting, conversational interop, or just allowing screen scraping. Who knows. HTI-5 said the information blocking stuff as proposed, where agentic access can be EHI exchange. HTI-6 would be some sort of certification requirement. So we don’t know the shape of it, but you get to choose a little bit. **Pryce Ancona (24:22):** Okay, well if this is a which would you rather, I definitely have my answer. But does anybody else want to go first? Brad? **Brad Thorson (24:28):** I’m ready. Ryan, are you ready? **Ryan Tucker (24:30):** I’m ready. **Brad Thorson (24:31):** All right. The most clear one for me, so I’m not going to go in order, but I’m entering matrimony with FHIR subscriptions. We need better signaling, and this brings our data model forward multiple decades. That’s the one we want to be with for a long time. If you disagree with me, you’re wrong. The complicated part is the other two, and I’ve got to say, I’m sorry, bulk FHIR, but you’ve got to go. **Brendan Keeler (24:57):** Aneesh Chopra is sad as he listens to this. **Brad Thorson (25:00):** Maybe not. If we’ve got agents that are able to vectorize that data, it doesn’t matter what format it’s in. Could we not just do a giant data dump into whatever your data model is and do some probabilistic estimation of I think this is an HbA1c number? If this is our decision set, then unfortunately, bulk FHIR, you are the weakest link. **Brendan Keeler (25:21):** Ryan, what do you got? **Ryan Tucker (25:22):** Maybe I’m going to cheat the game a little bit here, but I’m going similarly to Brad in my F. From Brendan’s description, agentic usage is a little flirty. You don’t know what it’s going to be, and frankly the next day you don’t know if you’re going to call it back. We’ll see what happens there. That would be my F. Marriage, I’m going to say bulk transactions. Usage for population, like can I actually set up my day for my whole organization. I’m not going to say bulk FHIR, because if we’re saying marriage, yeah, FHIR’s been around a while, but we’ve seen many. We’ve seen HL7 v2, v3, v4. I don’t know if FHIR is going to be around when I’m ninety. So bulk transactions is what I’m more interested in for marriage and seeing where that goes. And then I guess I’ve got to kill subscriptions, even though in the near term that’s probably what I want. I’d like to date subscriptions right now if I could, but I think that’s outside the rules of the game. So I’ll stop there. **Brad Thorson (26:29):** You’re wrong. **Brendan Keeler (26:30):** Pryce, what are we thinking? **Pryce Ancona (26:32):** What do I want the most? For sure subscriptions. Brendan, to your point, we’ve got these nearly ubiquitous nationwide networks and they’re query only. Everyone has to make a FHIR API available, query only. How is a new clinic going to hear about a patient? How is a new telehealth app going to hear about a patient? How is the remote patient monitoring cardiology program going to hear about a patient? It’s always an event triggered subscription, or an HL7 ADT, SIU, ORM. Those have been tried and true for us for longer than we’ve been alive. At least most of us. And sorry to the analysts listening who are like, I was implementing ADT in 1989. It’s time we have something in place that EHRs are forced to support, so that information blocking matters. If you can’t even figure out which patients you’re supposed to treat and why and when, then you have half the companies I know that do cool work running off manually copied Reporting Workbench reports to figure out who they should be treating. That’s just insane to me. So I’d want subscriptions to include upcoming encounters, encounters, registration, discharge, appointments, but also orders. If we don’t get orders, then I’m going to be like, what the hell is all this for? I don’t really like AI right now, just in general. Lots of mixed feelings. I saw some robots fighting today on LinkedIn and I was like, I don’t like that at all. So I’m going with bulk FHIR for number two. I also think it’s a nice stepping stone for all of these developers. Hey, now you could offer something that’s like a SQL search but for other people. You can understand that. That would be nice. And if we ask them all to implement MCP and we expect it to be as good as Databricks natural language queries, it’s just not. It’s going to be garbage. So I’m ready to stay away from the agency stuff for another couple of years. **Brendan Keeler (28:26):** I’m going to bend the rules too. I’m going to do a thruple with bulk FHIR and subscriptions. Let’s see the FHIR stuff through. Let’s see if this solves everything. Then we’re going to have to kill agents, because I don’t need Dario Amodei to get any more of my money. We don’t need data centers and we’re not ready. It’s where we started. We’re asking, well, what are we going to require? If we’re asking that, we’re probably not ready to put in requirements if we don’t know whether it’s MCP or what. So I think there’s potential for HTI-7 or 12 or something, but for now, burial for agents. With that, we’re at time. Thank you all. We’ll see you on the next one. Get full access to Health API Guy at healthapiguy.substack.com/subscribe