The Future of Physical Therapy and AI

Where AI is already useful

Most of what makes a PT clinic run well has nothing to do with clinical judgment — it's logistics. Answering the phone at 9pm. Verifying insurance before someone shows up for a visit they can't afford. Following up on a home exercise program three days after a visit instead of waiting two weeks for the next one. Chasing documentation so a claim doesn't sit in limbo.

That's the layer where I've actually deployed AI agents — Kai handles intake and scheduling, Nora handles insurance verification, Maya handles the calendar, Max handles billing and claims, Riley checks in daily on your home program. None of them decide what's wrong with you or what your treatment should be. They do the coordination work that used to eat hours of staff time, so more of my time goes to the parts of the job a person actually needs to be good at.

Where AI is genuinely changing the exam itself

This is the part I think is actually new, not just automation with a new name on it. Computer vision — the same kind of technology behind Atlas, the movement-tracking tool in this platform — can now watch how a joint moves through a phone camera and measure it. That used to require a motion lab, markers, and specialized equipment. Now it's a webcam and good software.

That doesn't make the exam smarter than a clinician. It makes it more objective. Instead of me eyeballing your squat and estimating "looks about 15 degrees short," the camera measures it, grades what it can defend, and greys out what it can't. I still interpret what that number means for you — the measurement just stops being a guess. If you want the deeper version of that argument, I wrote a whole book about the evidence problem in manual therapy: Cracked. This is the same instinct applied to measurement instead of technique.

Where it doesn't belong

None of the AI in this platform diagnoses you, and none of it decides your treatment. Not because the technology couldn't produce an answer — it's because a chatbot doesn't know your surgical history, hasn't watched you move in the room, and doesn't carry the liability if it's wrong. Every question our Smart Eval asks is neutral — "where does it hurt," "what makes it worse" — and every result goes to me, not to an algorithm making the call.

I've treated thousands of patients across university sports, professional athletes, and post-surgical recovery. Catching the case that doesn't fit the textbook — the referral that needs to happen today instead of next week, or spotting how your body's been working around the real problem — that comes from years in the room with people, not from a training set. AI can hand me better data faster. It can't replace having sat across from a patient and learned what their particular kind of "fine" actually means.

What I actually expect to change next

More objective measurement, not less — full-body range of motion, running and gait analysis, and eventually the full-body KinematiX Movement Screen (KMS), all measured the same honest way Atlas measures a squat today. More of the admin overhead disappearing so a visit is spent on treatment, not paperwork. What I don't expect to change: a licensed clinician reviewing your case, making the call, and being accountable for it. That's the part of "the future of physical therapy" that isn't actually up for debate, as far as I'm concerned.

See it instead of reading about it

The easiest way to understand where AI fits in this platform is to try the parts that use it. Take the free movement demo — a Smart Eval, a click-where-it-hurts picker, and a real Atlas scan, all built around the same rule: measure what the camera can defend, and let a person handle the rest.