Why the treatment room still needs a therapist
The question of whether radiation therapists will be replaced by AI runs into a simple fact: most of this job happens with a patient on a table and a beam about to switch on. A therapist positions the person to within millimeters, fits masks, boards and immobilization devices, lines up tattoos and lasers, then holds that setup steady for a daily fraction. Software can suggest a shift. It cannot ease a stiff shoulder into place or coach a nervous patient through a breath hold.
The second anchor is accountability during delivery. The therapist operates the linear accelerator, watches the patient on camera, stops the beam when someone coughs or moves, and checks the chart against the prescription before anything runs. Mistakes here are high-consequence and regulated, so hospitals keep a licensed person responsible for each fraction. Automation changes how the checks are done, not who signs for them.
Then there is the length of the course. Patients come back most weekdays for weeks. The therapist sees skin reactions, fatigue, weight loss and fear before anyone else, and passes that on to the oncologist and nursing team. Employment sits near 17,070 in the US with projected growth of 2.8% from 2025 to 2035, and median pay around $105,310 (BLS, 2025). That is a small, licensed workforce attached to expensive machines, which is not the shape of a job that empties out quickly.
What AI does, assists with, and leaves alone
Tasks our data marks as handled by AI make up 0% of task time. These are the paperwork and pattern jobs around the beam: drafting treatment records and dose logs, flagging mismatches between the plan and the delivered fraction, and first-pass contouring of structures that a dosimetrist or physicist then reviews. Our coverage score explained page sets out how that share is built; the headline coverage figure for this job is 12 out of 100.
Assisted tasks account for 19% of task time. Daily image matching is the clearest example: automated registration proposes the couch shift, and the therapist accepts, corrects or escalates it. Scheduling across machines, and checking a plan’s parameters against the prescription, also run faster with software doing the first look.
Work that stays with people is 81% of task time. Patient positioning and immobilization sit here, along with explaining the procedure and its side effects to patients and families, observing someone for a reaction during and after a fraction, and making the judgment call to halt treatment. None of that is a document task.
What the evidence actually tests
Most published work on AI in radiation oncology looks upstream of the therapist: auto-segmentation, treatment planning, adaptive replanning and quality assurance. There is no head-to-head test of a system doing a therapist’s console and table work against a qualified therapist over a full course of treatment. Our parity evidence grade for this job is D on an A to D scale, and a D means not measured, so we publish no parity number here.
What would settle it is specific: a prospective study across multiple centers comparing setup accuracy, imaging dose, treatment interruptions, error rates and patient-reported experience under automated setup versus standard therapist-led delivery. Until something like that exists, claims about AI matching therapists are untested. The quality parity method explains how a grade moves up, and the full scoring methodology covers the rest.
Good to know: an evidence grade of D is a statement about missing tests, not proof that AI performs badly at the task.
When this could change
Most likely after 2042 (8 in 10 of our scenarios). The replacement year method explains what that window is measuring and how it is drawn.
Two things could pull the date forward. Surface-guided and automated image-guided setup keep improving, and each step reduces the hands-on time per fraction. Adaptive planning at the machine is also maturing, which shifts work that once needed a therapist plus a dosimetrist into a shorter, more automated loop.
Two things hold it back. The physical side is the bigger one: our robotics panel rates the hardware needed for the hands-on tasks at the dexterous humanoid tier, which does not exist as reliable clinical equipment. The second is regulation and liability. State licensure, credentialing and radiation safety rules all name a responsible human operator, and the cost comparison on this page is between a software license and a trained therapist, not between a robot and a therapist. Treatment machines are also long-lived capital assets, so clinics replace workflows slowly.
How to stay needed in radiation therapy
Lean into the parts of the job that keep landing in the human column. Patient positioning and immobilization for difficult anatomy. Side-effect observation and escalation across a multi-week course. Explaining what is about to happen to a patient who is frightened, and to the family in the waiting room.
Two skills are worth building now. First, image-guidance judgment: knowing when an automated match is wrong and being able to say why. Second, quality assurance literacy, including how automated contours and plan checks fail, so you can catch the failure rather than approve it. Therapists who can document and challenge an automated output are the ones clinics keep near the console.
If you are weighing adjacent paths, the closest work sits with medical dosimetrists, who build the plans therapists deliver, and with radiologic technologists and nuclear medicine technologists on the imaging side. You can also see how this role sits inside healthcare practitioner jobs and the wider healthcare sector.
Compare two jobs side by side if you are choosing between them, or scan the jobs that mostly need a person list to see where patient-facing clinical work lands overall.