Trajectory of Artificial Intelligence (AI) in Medicine
Started by DrMetal
why not? But ok, say you don't get "laid off" in the formal sense. If you're a 1099 independent contractor, as many of us are, it's very easy to let your 'contract' expire and not renew (unless you agree to new terms). If you are an employed W2, now your boss will make you see more volume, take on a couple mid-levels, etc. As older physicians leave the group, its just easier not to fill their spots (the hospitalist group of 20 will dwindle down to 10, maybe even 5). There's a lot of ways the industry can mess with us.
I don't know that physicians are necessarily advocating for it . . . but we're recognizing that it's here. The train has arrived, it's in practice. There's really no turning back.
Protecting patients? [sarcasm/] Yeah, we've been doing that great for the last 25 years. [/sarcasm]
Yeah no disagreement from me there. I’m a 1099 myself and know my position is only as safe as I keep providing value in this rapidly changing landscape. I, too, have had to use AI bc the bar of expectations to meet is higher and I simply can’t just sit there crafting MDM narratives from scratch. Being really good at prompting and then reading/reviewing efficiently has been way more of an ROI. Templates and smartphrases aren’t even sufficient anymore bc the notes risk looking cloned.
The w2’s are a bit more protected bc of optics but their employers will simply make them run faster on the hamster wheel. You’re already seeing this play out especially with new psychiatry and hospitalist attendings. Jobs seem to be trending towards slim pickings.
And @chessknt I am not advocating for it. More just acknowledging what’s already happening and will accelerate going forwards.
Last edited:
Advertisement - Members don't see this ad
I mean I don't know what you do typically but this last week alone I prevent an unnecessary VATS by a terrible surgeon by forcing myself in to an empyema case, forced cardiology to stop gaslighting someone and acknowledge the heart actually isn't fine with a elevated ve/vco2 on a cpx, put someone back on actual therapy for asthma instead of the inhaled bull**** their naturopath was giving them, and finally got an anti-IL5 approved through the obstructionist insurance for chronic fibrosing eosinophilic lung disease in a lady dying of steroid poisoning and that is just off the top of my head.
I know the above happens all the time all around the country. I am sure you have done it too even if you don't think of it that way but we all advocate for and protect patients against other crappy doctors insurance companies, and sometimes even themselves.
My big concern with all this is that AI again won’t actually make this shorter. Until it can actually carry out a whole visit by itself, it’s not going to shorten the process. And so in the relatively near term, I can see admin telling everyone “look guys, you need to see 40-60 pts a day, you have AI”. Except that it won’t actually make the bulk of the visit faster, and pts will be pissed with 5 min visits. And then admin will say “doctor it’s your responsibility to keep the satisfaction scores high too”, despite 5 min visits that nobody likes.
I think there will be a solid cash pay market for doctors willing to give an AI free, cash only “classic medicine”experience to pts who want it (and can afford it). Everyone else will get ****tier healthcare. Remember Blade Runner? It was an expensive status symbol to have a real live pet. Everyone else had a robot…
On the flip side: there are opportunities in any situation. And so the real question is…what can you do to come out “on top” in what’s coming for medicine?
Last edited:
I'm a general internist/Hospitalist, so my job is to protect patients from themselves.
I can't wait to see AI do an interview with one of my older patients for which every yes/no question is an incredible opportunity for a story.
Or for the patient with a panpositive review of systems.
These are the fails I live for online.
Or for the patient with a panpositive review of systems.
These are the fails I live for online.
There's no storytelling when there's no humans involved. True with respect to the practice of medicine, and bartending . . .
if #ROSPositive >=5 {
Psych Consult ;
}
Easy day.
I think the order and degree of AI disruption is going to be very uneven by specialty.
The first and most exposed groups are probably the non-patient-facing specialties. The big two are diagnostic radiology and pathology, with DR probably #1. AI advancement in imaging is much farther ahead than almost anywhere else in clinical medicine right now. I do not think AI is going to completely replace radiologists or pathologists anytime soon, but I do think it will change their labor economics the most, and the soonest.
Radiology is a good example even if looking back historically. Over the last 20 years, technology has allowed radiologists to read much higher volumes, and CMS has already responded over time by reducing inflation-adjusted reimbursement for imaging. Radiology compensation is still very high and the job market is excellent right now because imaging volume has exploded and AI is still mostly augmenting radiologists rather than replacing them, though this will only be temporary. But if AI meaningfully increases reads per radiologist again, I would expect CMS/payers to eventually respond the same way: again lower reimbursement per unit of work. Once that happens, the number of radiologists needed by the market could become much lower than the supply produced by resident graduation minus retirements/attrition. This will make it harder for especially new grads to land a desirable job. IR is obviously much more protected from this than DR.
Patient-facing specialties ,whether office or shift-based, are going to be much harder to replace because the actual patient encounter is still a major bottleneck. The hard part is not just generating the note or suggesting an order set. It is getting a usable history from a real person, dealing with family, nonadherence, poor health literacy, social issues, unrealistic expectations, and all the messy human stuff. AI will still mainly augment these specialties) and can help with chart review, notes, MDM wording, discharge summaries, inbox work, and maybe protocolized decision support, but it does not magically make a 15-minute human encounter into a safe 5-minute encounter.
That said, patient-facing specialties are not immune. If AI automates more of the EHR/charting burden, I can absolutely see CMS and payers eventually arguing that RVU values should fall relative to inflation because the “work” is easier. The response from employers would be predictable: see more patients to maintain the same income. I suspect there will be a longer lag for this in patient-facing fields than in radiology/pathology.
Procedural specialties are probably safest for now, although not completely safe long-term. Robotics and AI will continue to augment procedures, and eventually some procedural work may be standardized or delegated more than physicians expect. But replacing the hands-on proceduralist is a much harder problem than replacing part of the cognitive/documentation workflow.
In terms of APP leverage, a experienced PA/NP with 5+ years in a specialty who is very good at using AI could become much more valuable to a health system. Not the same as a strong attending, obviously, but for routine specialty-specific workflows, their practical output may start looking closer to a junior attending than many physicians are comfortable admitting. From an employer perspective, that is going to matter if they can still pay them much lower than a junior attending as they do now.
So my view is not “AI replaces all doctors.” It is more that AI compresses the economic value of certain physician tasks. The physicians who come out ahead will be the ones who use AI well (ie much more advanced than someone just knows the basics of prompting Chatgpt or Claude, with skillsets probably closer to AI/ML engineer), understand its limits, and can prove value beyond just generating notes and clicking orders. The ones who ignore it or treat it as a fad are probably going to be in a weaker position (probably won't matter as much for older physicians who are about to retire anyways, but obviously a bigger issue for new and recent grads who still have most of their career ahead). .
The first and most exposed groups are probably the non-patient-facing specialties. The big two are diagnostic radiology and pathology, with DR probably #1. AI advancement in imaging is much farther ahead than almost anywhere else in clinical medicine right now. I do not think AI is going to completely replace radiologists or pathologists anytime soon, but I do think it will change their labor economics the most, and the soonest.
Radiology is a good example even if looking back historically. Over the last 20 years, technology has allowed radiologists to read much higher volumes, and CMS has already responded over time by reducing inflation-adjusted reimbursement for imaging. Radiology compensation is still very high and the job market is excellent right now because imaging volume has exploded and AI is still mostly augmenting radiologists rather than replacing them, though this will only be temporary. But if AI meaningfully increases reads per radiologist again, I would expect CMS/payers to eventually respond the same way: again lower reimbursement per unit of work. Once that happens, the number of radiologists needed by the market could become much lower than the supply produced by resident graduation minus retirements/attrition. This will make it harder for especially new grads to land a desirable job. IR is obviously much more protected from this than DR.
Patient-facing specialties ,whether office or shift-based, are going to be much harder to replace because the actual patient encounter is still a major bottleneck. The hard part is not just generating the note or suggesting an order set. It is getting a usable history from a real person, dealing with family, nonadherence, poor health literacy, social issues, unrealistic expectations, and all the messy human stuff. AI will still mainly augment these specialties) and can help with chart review, notes, MDM wording, discharge summaries, inbox work, and maybe protocolized decision support, but it does not magically make a 15-minute human encounter into a safe 5-minute encounter.
That said, patient-facing specialties are not immune. If AI automates more of the EHR/charting burden, I can absolutely see CMS and payers eventually arguing that RVU values should fall relative to inflation because the “work” is easier. The response from employers would be predictable: see more patients to maintain the same income. I suspect there will be a longer lag for this in patient-facing fields than in radiology/pathology.
Procedural specialties are probably safest for now, although not completely safe long-term. Robotics and AI will continue to augment procedures, and eventually some procedural work may be standardized or delegated more than physicians expect. But replacing the hands-on proceduralist is a much harder problem than replacing part of the cognitive/documentation workflow.
In terms of APP leverage, a experienced PA/NP with 5+ years in a specialty who is very good at using AI could become much more valuable to a health system. Not the same as a strong attending, obviously, but for routine specialty-specific workflows, their practical output may start looking closer to a junior attending than many physicians are comfortable admitting. From an employer perspective, that is going to matter if they can still pay them much lower than a junior attending as they do now.
So my view is not “AI replaces all doctors.” It is more that AI compresses the economic value of certain physician tasks. The physicians who come out ahead will be the ones who use AI well (ie much more advanced than someone just knows the basics of prompting Chatgpt or Claude, with skillsets probably closer to AI/ML engineer), understand its limits, and can prove value beyond just generating notes and clicking orders. The ones who ignore it or treat it as a fad are probably going to be in a weaker position (probably won't matter as much for older physicians who are about to retire anyways, but obviously a bigger issue for new and recent grads who still have most of their career ahead). .
Yup. It’s an arms race and already is.
For 2026, Congress authorized a temporary payment bump, raising the baseline Conversion Factor to roughly $33.40. On paper, it looks like a raise. However, CMS immediately neutralized this for many specialties by introducing a permanent 2.5% "Efficiency Adjustment" that cuts the work RVUs for non-time-based codes. The government's premise is that over time, doctors get faster at diagnosing and performing procedures due to workflow and tech improvements, so Medicare should pay less for that work. They give with the statutory update, and they take with the RVU math.
CMS is intentionally making the traditional FFS treadmill uncomfortable to force physicians into Alternative Payment Models (APMs) and value-based care. The 2026 rule explicitly splits the fee schedule: physicians in qualifying APMs receive a higher Conversion Factor ($33.57) than those who remain entirely FFS ($33.40).
When the Part A trust fund approaches its depletion date (currently projected around 2036, though previously 2033), Congress will not let a catastrophic, automatic double-digit cut happen overnight. The political fallout of bankrupting community hospitals would be too severe. Instead, the government will use a "death by a thousand cuts" approach—leveraging obscure, highly technical policy levers to slowly squeeze money out of the system.
Because Part B is funded by premiums and general tax revenue, it technically cannot go bankrupt. But because it continually demands more from the federal balance sheet, it is subject to intense, highly targeted cost-containment measures. The short answer for physician income is: you have to run faster just to stay in the same place.
This was happening before AI. If they didn't call it an efficiency adjustment they would have just named it some other bull**** and done the same thing. When the RUC starts to kill wRVU for E/M coding that will be GG to non-procedural medicine.
