What changes for physical therapy with AI?

The shift

Designing a rehab program, tracking progress, and analyzing movement from video go from scarce (a licensed therapist's clinical time and trained eye) to abundant and near-free: an app builds a progressive exercise plan for a diagnosis, logs adherence and pain scores, and reads range-of-motion or gait from a phone camera. What stays scarce is the hands, the physical assessment, and the accountable clinician who owns the risk of acting on all of it.

The axioms

• The therapist's value is prescribing the right exercises for the injury and progressing them over the course of care — scarce clinical knowledge.
• Tracking whether the patient is improving — range of motion, strength, pain, function — takes a trained observer measuring at each visit — scarce assessment time.
• Reading how a body moves — gait, compensation patterns, a faulty movement — takes a clinician's trained eye in the room — scarce observational skill.
• Manual therapy — mobilizing a joint, releasing soft tissue, hands-on techniques — requires a person's hands on the patient — scarce physical action.
• Physical assessment — palpation, special tests, feeling end-range and tissue quality — requires a clinician touching and moving the patient — scarce hands-on judgment.
• A licensed clinician must own the diagnosis, the plan, and the decision that a given load is safe for this patient today — scarce accountability under a license.
• Patients don't reliably do their home exercises alone, past the point pain eases and motivation fades — scarce adherence.
• Care is billed per visit, with assessment, plan, and hands-on treatment bundled into the session fee — a pricing habit resting on scarce clinical time.
• Recovery requires the patient to actually load the tissue, correctly and repeatedly, in a body — scarce physical action by the patient.

Invalid axioms

1. The therapist's value is prescribing and progressing the exercise program. For common musculoskeletal presentations, the exercise prescription is well-documented, protocol-driven knowledge — an app generates a sound progressive plan for a diagnosed condition, adapts it to logged pain and reps, and explains the reasoning. Habit-trap: clinics still position "your custom program" and the printed exercise sheet as the deliverable, when that half is now the cheap, commoditized part of care.
2. Progress tracking needs a trained observer at each visit. Logging range of motion, pain scores, adherence, and function — and flagging when a trend stalls — is exactly what an app does continuously and for free, and it captures the six days a week the patient isn't in the clinic. Habit-trap: re-assessment gets treated as billable clinician work when the routine measurement and charting is now near-free and better sampled at home.
3. Reading movement takes a clinician's eye in the room. Phone-camera motion analysis of gait, squat depth, or shoulder range is good enough for many exercise-supervision and monitoring purposes, at zero marginal cost and every rep rather than a sampled few. Habit-trap: clinics bill observation of unloaded, low-stakes movement as skilled time when a camera now covers much of it. (This is the fastest-moving call here — markerless pose estimation is improving quickly. It is credible for range and rep-quality monitoring; it is not yet a substitute for assessment of loaded, near-limit, or acute-injury movement, and confidently-wrong readings are the failure mode.)

Unchanged axioms

1. Manual therapy is hands on a patient, and can't be produced as tokens. Mobilizing a stiff joint, handling an acute post-op limb, guiding a movement with contact — none of this is generation. Whatever share of PT outcomes comes from hands-on work stays gated by a person in the room, and no amount of abundant planning touches it.
2. Physical assessment requires touching and moving the patient. Palpation, special tests, feeling tissue quality and end-range, sensing guarding or apprehension — the clinician's hands are the instrument, and the input doesn't exist for a model that can only see and read. An app can screen and triage; it can't feel a joint.
3. A licensed, accountable clinician must own the plan and the decision to load. Whether this patient's tissue tolerates this load today, when to progress, when to stop, when something is a red flag needing referral — this is judgment under stakes, carried under a license a model can't hold. Being confidently wrong about loading a healing tissue or missing a serious pathology is more dangerous than almost anywhere the plan is free.
4. Someone still has to make the patient do the work, and that someone is human. The scarce thing was never the plan — it's execution over weeks against pain, fear of movement, and fading motivation. A notification isn't accountability; a person expecting you, adjusting when you're discouraged, and catching you before you quit is. Abundance in planning does nothing to the fact that most patients don't sustain rehab alone.
5. In-person presence keeps the work safe and calibrated in real time. Reading a patient's energy, apprehension, and pain response mid-session and pushing or backing off accordingly is real-time judgment on a body, not a text output. It stays gated by a clinician being present.

New axioms

1. When the plan and the tracking are free, clinics have to price presence, hands, and accountability explicitly — and there's no convention for it. Billing was built around the visit as the unit, with the plan and re-assessment bundled in. "Pay me for the hands-on work," "pay me to own the risk of progressing you," and "pay me to make you actually do this" are the real products now, but insurance codes, per-visit norms, and patient expectations all still assume the bundle. The clinic that can't name what it's charging for will keep billing the free part.
2. Patients will act on confident-wrong AI motion analysis and self-prescription. A patient with a phone app gets a plausible movement reading and a plausible program and starts loading — with no clinician to catch that the reading missed a compensation, the plan is wrong for their actual pathology, or their symptoms are a red flag. The abundance produces plausible clinical output with no accountability attached, and someone has to solve for who catches the miss before the patient is hurt.
3. Verifying AI assessment becomes the scarce clinical act, and it has no workflow. If plans, tracking, and motion reads are abundant, the defensible clinician work is checking them against the body in front of them — confirming the camera's gait read against palpation and history, sanity-checking the app's progression against how the tissue actually responds. But there's no established way to verify AI clinical output at volume, and nobody owns the review step.
4. Adherence the app can't drive still gates outcomes — and now has to be sold as the product, not thrown in. The home-exercise app makes prescription free and visibly competent, which can read to the patient as proof they don't need a clinician — missing that the thing they're bad at, doing the work for weeks, is the thing the app doesn't solve. Clinics have to make the case for a value the patient can't yet see they need, against free evidence that seems to say otherwise, and there's no accepted measure of a clinician's adherence lift.

Where it breaks

Clinics keep pricing and billing the plan and the re-assessment — the custom program, the tracked measurements, the observed exercise (INVALID) — while the only parts a patient can't get free are the hands, the physical assessment, and the accountable decision to load (STILL HOLDS), and those have no pricing convention separate from the visit bundle. The clinic is charging for the commodity and giving away the scarce thing, and neither insurance coding nor patient expectation has been rebuilt to let it do the reverse.

Separately: the patient can now get a free app that writes a competent program and reads their movement (INVALID: the plan and the tracking were the visible value), which reads as proof they don't need a therapist — right where the app is most likely to be confidently wrong about a body it can't touch and can't be liable for (NEW: verification and accountability are the scarce acts). The evidence the patient can see points them toward self-treating exactly when the missing clinician matters most.

ai-axioms.vercel.app/axioms/healthcare/physical-therapy-ai-axioms