will AI replace radiation therapists?

Every single task in your role sits at 0% AI penetration. That's not a rounding error. It means that across 22 separate tasks analysed by O*NET, not one of them is being handled or even meaningfully assisted by AI at scale today. That's rare, and it matters.

The reason is straightforward. Your job is built on things AI structurally can't do. You position patients on treatment tables with precision, adjusting for each person's anatomy, pain tolerance, and anxiety level in real time. You read a patient's face when they flinch. You decide whether that flinch means you should pause, adjust, or reassure. A machine can deliver radiation. Only you can decide what that moment requires. Positioning accuracy alone, where a millimetre off-target can mean healthy tissue gets irradiated, demands the kind of embodied spatial judgment that robotics researchers have been chasing for decades without cracking it in clinical settings.

You also carry legal and clinical accountability. You review the prescription, confirm patient identity, check the diagnosis, and verify the treatment plan before any dose is delivered. If something's wrong, you're the one who catches it and acts. According to BLS data, radiation therapists work under physician direction but conduct most treatment sessions independently. That independence is a signal of professional judgment, not just technical execution. AI can process a treatment plan. It can't sign off on it, adapt it in the room, or take responsibility for what happens next.

AI systems trained on medical imaging and treatment planning data can suggest dose distributions or flag anatomical structures on a scan. But suggesting is not the same as treating. The gap between a treatment plan on a screen and a patient on a table is where your entire job lives, and AI has no meaningful presence there.

Hallucination is a real and documented problem in clinical AI. Systems like GPT-4 and similar large language models have produced fabricated drug dosages, invented citations, and generated plausible-sounding but wrong clinical information. In radiation oncology, where dose precision is measured in centigray and errors can cause radiation necrosis or treatment failure, a system that confidently produces wrong outputs isn't a productivity tool. It's a liability. No hospital is handing dose administration decisions to an AI system that can't be held accountable when it's wrong.

There's also no regulatory path yet for AI to operate independently in direct patient treatment. The FDA has cleared AI tools for imaging analysis and treatment planning support, but those are physician-facing tools that feed into your workflow upstream. The moment of treatment itself, where you're watching the patient, checking equipment function, and managing unexpected reactions, has no AI equivalent on the market. The human in the room isn't optional. It's a licensing requirement.

To be honest with you: the AI penetration in your specific role is effectively zero right now. But AI is active in the broader radiation oncology workflow, just upstream of you, in areas your colleagues and the physicians you work with are starting to use.

Treatment planning software like Varian's Eclipse and Elekta's Monaco have added AI-assisted dose optimisation features. These tools help medical physicists and dosimetrists model dose distributions faster, sometimes cutting planning time from hours to minutes. That's real. But that work happens before you ever see the patient. By the time someone is on your table, the plan is set. You're the one who delivers it.

On the documentation and scheduling side, some cancer centres are piloting ambient AI scribes and AI-assisted scheduling tools to reduce administrative load on clinical staff broadly. Tools like Nuance DAX are used in oncology departments for physician notes. You might see these tools touch your workflow at the edges, in shift handover notes or treatment summaries. But your core tasks, positioning, dose delivery, equipment checks, patient observation, and real-time reaction management, have no AI product actively operating in that space today. The vendors aren't there yet, and the clinical and regulatory barriers are high enough that meaningful AI involvement in direct treatment is years away at minimum.

Your day hasn't changed much, and that's the honest answer. The rhythm of your work, patient in, position, verify, treat, document, patient out, is essentially the same as it was five years ago. You're not spending less time on any core task because of AI. There's no documentation tool running in the background drafting your treatment notes while you work.

What you might notice is upstream. If your department uses AI-assisted planning software, the plans arriving in your queue may be more refined, sometimes with tighter margins already modelled. That doesn't change what you do, but it might mean fewer mid-treatment plan adjustments flagged back to the physicist. The plan is better before it reaches you.

What hasn't changed at all is the patient-facing part of the job. You still spend the bulk of your time in the treatment room. You still talk people through their fear. You still troubleshoot equipment that behaves unexpectedly. You still make the call to pause when something looks wrong. That's not changing because of AI. It's changing, if at all, because of shifts in treatment technology like stereotactic body radiation therapy and proton therapy, which require different positioning precision and patient management skills, not because a software tool is taking anything off your plate.

The BLS projects 1.9% job growth for radiation therapists between 2024 and 2034. That sounds modest, but it's growth in a field with only 19,200 people employed and roughly 900 openings per year. A small absolute number of jobs in a small workforce still means steady demand.

The growth isn't AI-driven. It's cancer-driven. The American Cancer Society estimates that cancer diagnoses in the US will exceed 2 million annually by the mid-2030s, driven partly by an ageing population. More cancer patients means more demand for radiation therapy, regardless of what AI is doing in adjacent parts of healthcare. The Anthropic Economic Index ranks this role among the lowest AI exposure occupations in healthcare, which means it's not a job where AI is quietly absorbing work and making fewer humans necessary.

There's also no offshoring risk here. You can't deliver radiation therapy remotely. The job requires a licensed professional in the room with the patient, every time. That physical requirement, combined with genuine growth in the patient population you serve, puts this role in a more stable position than the flat growth number alone suggests. A 1.9% growth rate in a role with zero AI displacement pressure is genuinely more secure than a 10% growth rate in a role where AI is eating half the tasks.

Your exposure score is 0% today and it's likely to stay very low for at least the next 10 years. For that to change meaningfully, we'd need AI systems that can physically position patients, read real-time physiological and behavioural cues in a treatment room, and hold legal accountability for clinical decisions. None of those capabilities are close. Robotics in clinical settings has advanced, but autonomous patient handling in a radiation therapy context is nowhere near regulatory approval or practical deployment.

The area to watch is AI in treatment planning and adaptive radiation therapy, where systems like Varian's Ethos already adjust plans between fractions based on daily imaging. If adaptive therapy becomes the standard of care, your role may shift toward more real-time plan review and less reliance on a fixed pre-set plan. That's a change in workflow complexity, not a reduction in the need for you. If anything, more sophisticated treatment technology has historically increased the skill requirements for radiation therapists, not replaced them.

The best thing you can do right now is go deeper into the clinical and technical complexity of your role, not broader. Radiation therapy is splitting into subspecialties that require distinct skill sets: stereotactic radiosurgery, proton therapy, MR-guided radiation therapy, and adaptive treatment planning. Each of these commands higher pay and is harder to fill. If your department offers cross-training in any of these areas, take it.

Get comfortable with the treatment planning systems your department uses, even if plan creation isn't your primary job. Understanding how Varian's Eclipse or Elekta's Monaco works, and how AI-assisted optimisation is changing what physicists hand you, makes you a more informed practitioner. You'll catch plan anomalies faster. You'll communicate better with dosimetrists. That upstream literacy is worth building.

On the patient side, double down on the communication and observation skills that your 0% exposure score reflects. Anxiety management, patient education about side effects, and the ability to keep someone calm and still through a 20-minute treatment session are skills that matter clinically and that no tool replaces. Some cancer centres are also expanding radiation therapists into roles that include patient education coordination and treatment review support. If your organisation is moving that direction, those are worth pursuing. The career risk for radiation therapists isn't replacement. It's staying too narrowly defined when the role is quietly expanding.