If you’re a radiology resident scrolling anxious posts about AI replacing reads and CMS cutting reimbursement, here’s the uncomfortable part: both things you’re worried about are real. And here’s the part that should actually change how you weigh them — they’re not pointing in the direction most of the anxiety assumes.

Start with what’s actually happening to reimbursement

CMS finalized its 2026 Physician Fee Schedule with a modest overall conversion factor increase, but the details work against imaging specifically: diagnostic radiology is facing roughly a 2% net cut once practice-expense and efficiency adjustments are factored in, alongside similar pressure on nuclear medicine and radiation oncology. Interventional radiology fared better, gaining ground on several fronts. This is a real, structural squeeze — not a rumor — and it’s worth knowing precisely because vague anxiety (“reimbursement is getting cut”) is less useful than the actual shape of the cut, which varies significantly by subspecialty within radiology itself.

That distinction matters for fellowship choice more than the topline number does. A cut concentrated in facility-based diagnostic work affects a hospital-employed general radiologist differently than it affects an interventional radiologist or someone in a subspecialty with genuine supply constraints.

Now the AI picture — which cuts against the doomer narrative more than residents often expect

The most consistently repeated fear is that AI is coming for radiology jobs specifically. The actual labor market data tells a different, more complicated story. Radiologist demand has continued growing even as AI tools have become deeply embedded in imaging workflows — roughly half of U.S. radiologist job openings went unfilled in recent years, and workforce shortfall projections extend out for decades, driven by imaging volume growth that’s outpacing the residency pipeline. Radiology residency programs are filling at or near capacity, which is not what you’d expect to see if incoming trainees broadly believed the specialty was being automated away.

The mechanism matters here: AI in radiology so far has mostly increased throughput and workflow efficiency rather than reducing the number of radiologists needed, because imaging demand itself has been growing faster than efficiency gains can offset. That’s a different dynamic than a technology simply substituting for the job. It also means the effect isn’t uniform — subspecialties and geographies with more constrained training pipelines and higher complexity (pediatric and emergency radiology, for instance) have shown stronger, faster-filling demand than more commoditized, higher-volume segments, which are also the segments more exposed to automation pressure. The generic version of “AI will affect radiology” obscures a real and useful distinction between which parts of the field are more and less exposed.

Why this matters for fellowship and career decisions right now

The mistake isn’t taking AI or reimbursement trends seriously — it’s treating them as a single undifferentiated threat to “the specialty” rather than researching how they actually land on the specific subspecialty and practice setting you’re considering. A few more useful questions than “is AI going to replace radiologists”:

• Which subspecialties within my field have genuine training-pipeline constraints, and which are more commoditized? Constrained pipelines are more insulated from both automation pressure and oversupply, almost by definition.
• How exposed is this specific reimbursement pathway to the kind of cuts I’m seeing — facility-based versus office-based, procedural versus interpretive? The CMS cuts landing on radiology in 2026 aren’t uniform across the specialty, and the structural difference (facility versus non-facility practice expense treatment) will likely keep mattering in future rulemaking cycles too.
• What does the actual current job market show, not the anxious framing of it? Job-openings data, fill times, and compensation trends are observable now — they’re a better guide than extrapolating from a worst-case AI headline.

The broader point beyond radiology

Every specialty residents are choosing between has some version of this dynamic: real regulatory and technology pressure, combined with real structural demand factors that don’t move in lockstep with the pressure. CMS reimbursement changes happen most years, to some specialty, for reasons that are usually more about payment methodology than about the value or future of the specialty itself. AI is being integrated into nearly every corner of medicine, and the pattern so far — across radiology and elsewhere — has more often been augmentation of existing physician capacity than replacement of it, though that pattern is not a guarantee about the more distant future.

None of this means blind optimism is the right response either. Reimbursement pressure is real and worth factoring into practice-setting decisions (a facility-based, hospital-employed diagnostic role and an outpatient interventional practice are exposed differently to the specific cuts happening right now). But making a fellowship decision off of headline-level AI anxiety, without looking at what the actual job market and reimbursement data show for the specific niche you’re considering, is a worse decision-making process than the fear itself.

You don’t have to resolve every uncertainty about the next twenty years of medicine before choosing a fellowship. You do owe yourself a look at the actual current data on your specific subspecialty before letting a scary headline make the decision for you.