Nine months ago, on a routine 13 mile run, I noticed a strange dull ache in my left knee.

I had run a lot that week, so I ignored it as routine soreness. But by the end of the run I could barely walk.

Weeks passed, then months, and I still couldn’t walk without pain. I met with three doctors and a physical therapist and they all deemed it “patellar tendonitis.” Some said to “just quit running,” while others suggested complex therapies my insurance wouldn’t cover.

I felt hopeless. I was drowning in conflicting blog posts and confusing medical papers, with no clear direction.

I’m not 100% back today, but I feel close. I can walk without pain, and just last week I ran two miles for the first time since the injury. It’s crazy, but the thing that made the difference for me was using a team of AIs as my physical therapists.

Understanding the Injury

December 2024. It’s 1am and I’m 12 replies deep into r/kneeinjuries. I have no idea what these people are talking about.

Eccentric loading. KOT. VISA-P scores. I couldn’t even understand the injury, much less treat it.

I had a feeling AI could help, but I wasn’t sure. I tried ChatGPT, but wasn’t super confident in what it was saying.

Coincidentally, right around that time my two friends Charlie and Jackson had just released a chat app called Chorus. I figured I’d test out my use case on it.

Chorus let me see multiple responses from different models side by side, which actually made me far more trusting in the results. It felt like getting a second opinion from a doctor before a surgery.

Here’s an example: in physical therapy, there’s a framework called the “pain scale.” You use it to rate your pain from 1-10 and monitor progress. I wanted help figuring out how on earth I was supposed to figure out if my ache was a 3 or a 7:

Above, Claude gives me concrete benchmarks to evaluate my pain (a 3/10 feels like a “warming sensation” during a decline squat) while Gemini emphasizes comparing pain across days (”next-day pain response” is the “most critical metric”).

Like a team of doctors, my team in Chorus — Opus 4, Gemini 2.5 Flash, and GPT-4.1 — gives me a broader picture than any individual could.

Down the Rabbit Hole

I felt like I was finally starting to get patellar tendonitis. But no one seemed to agree on how to fix it. NSAIDs? No NSAIDs. Rest? No rest. Bodyweight squats? Heavy squats.

I needed to go deeper. So I went to YouTube.

Thank god for the algorithm, which led me to the “Patellar Tendon King,” (although now he does other tendons too), Jake Tuura.

Jake’s videos linked me to studies on treating patellar tendonitis, which I fed to Chorus. Maybe it could synthesize these papers for me?

And it worked…surprisingly well! Recent research pointed me in a clear direction: using heavy isometrics as an anchor to begin rebuilding my knee’s strength and tolerance to load. And most importantly, I felt like I could finally see why.

Engineering My Comeback

Armed with Chorus, Jake Tuura, and a bloodstream free of painkillers (turns out they don’t help long-term tendonitis), I felt confident to make a real training plan.

I summarized my most important chats, and asked my chorus of models to build me a multi-month recovery plan, with checkpoints and pain gauges along the way.

Unfortunately, I’m a chronic worrier.

I naturally had many follow-ups to ask: what does a “3” on the pain scale really mean? Should I do single leg wall sits and Spanish squats? Or choose just one!?

As my chats grew, the models started to get more confused. They’d forget things I’d previously mentioned about my condition, or hallucinate parts of my training plan. I had to do a lot of manual clarification, branching, and copy pasting of context.

Relatedly, I recently joined Chorus and helped build Projects, which gives my chats a shared context about my medical situation. It also automatically summarizes each paper and question I’d previously discussed with the AI so each chat knows the latest about my knee.

Getting Stronger

The interesting part about this experience was that I still needed the expertise of a doctor to get a diagnosis. But AI filled a critical gap in helping me understand my injury.

There’s a lot of talk about AI replacing X job, and a naive takeaway from my experience is that AI could replace my doctor. Instead, it felt like the AI augmented my medical experience by filling a bunch of open holes.

AI was always on-call to teach me the simple stuff. It was someone who could field my dumb questions, or tweak my training program for the week based on how I’m feeling. It helped automate the mundane, day to day parts of medical care that experienced doctors don’t need to be doing. AI helped me internalize and understand what’s wrong with my body and feel confident that things will get better.

I’m not back to 100%, but I’m further along than I’ve ever been. And, I just ran two miles last week for the first time in nearly a year. Next week I’m running three.