ACL

August 2026

I fell off a cliff skiing, tore my ACL, had a knee surgery, then built an AI agent that phone calls me several times and yells at me. (If I don't log recovery exercises)

I called it Bars and it now:

Tells me how I slept reading my Oura data by itself each morning
Phone calls me twice a day to ask if I did my recovery exercises and if I hit my protein norm (interrogates me otherwise)
Reads my Granola summaries to reason through notes from my surgery team and physios
Emails weekly reports on my progress to my physio and can enquire on appointments with my clinic
Prepares daily summary of upcoming Luma events in London
Sends tweets I bookmarked to my Kindle
Launches a Codex session through a phone call and onboards to apps through Stripe Projects

It's just been blowing my mind. It's a new form factor that's hard to explain verbally. Here is most tangible analogy. Imagine the apps you use daily and devices/items you interact with could collectively tell you about your cognition and physiology. It's digital and physical worlds colliding to work for you specifically.

It's not clear what it should look like. Maybe a dashboard, home assistant, voice agent, robot or a pocket device. But something might be coming to consumer AI.

In my blogpost (maybe turns into a series) I talk about how I went through ACL surgery and through prism of recovery explored the limits of personal agents.

🔏 no LLM was used in writing a single character in this piece.

ACL injury fucking sucks. It's the worst. It was my 9/11. I'm not surprised athletes cry right on the pitch when they get their ACL ripped apart. You are basically stripped off of your favourite activities for half a year minimum. It took me a while to process that it happened.

ACL surgery itself is a bit nuts. They get a tendon from another part of your leg, create a new handmade ligament, drill through a few bones in your knee and thread through a newly created ligament and hang it. It imitates the original ACL.

As any engineer I started overthinking every aspect of ACL prepping for surgery. I got absorbed. At this point I didn't know the absolute lowest low is a post surgery period, but first let me talk through how I prepared.

Pre surgery

It started with me debating ChatGPT on what ACL graft (tendon from different part of the leg) to choose, pros/cons of donor ACL and how latest ACL research literature is aligned with clinics in London specialising in knee surgeries.

Yeah, it's not like choosing sneakers. You buying yourself a damn knee.

You get overwhelmed with hundreds of little things that you have to think through. Worst thing is every small thing codepends on each other. Below is just the tip of the iceberg:

To have a surgery or just leave it torn?
What surgeon to have a surgery with? Does he have enough experience and what clinic does he work for?
How to even choose the right surgeon and clinic?
Is the clinic and surgeon covered by insurance? What is the claim procedure, what does it cover?
What procedures does the surgery involve, is it all covered by the policy?
What graft to choose? Is it applicable to patient goals and background
To have a LET or not? What are the consequences of having it vs not? What does literature say about it?
To use a donor graft or your own tendon, pros vs cons
When to have a surgery? How long recovery takes depending on surgery choice?
How to plan my vacation time and logistics after the surgery?
What physiotherapist to go for? Better closer to home, closer to office or just whatever is best

You can ask all questions in a single chat and pray for compaction algorithms not to wipe out valuable information about your knee MRI scans results as the chat goes longer and longer. Spoiler: it won't remember all details of what happened to you in the 4 question just because of how much each question can branch out. Alternative, have a separate new chat for each question. But then you will have to explain everything afresh every single time. It's like having to go to a new barber every week. It'll most definitely suck and no one is to blame. Such is life.

Last resort is you might turn on the memory feature. But what if some day you want to try a different platform model that is potentially much much smarter than what you have on this platform you got glued to?

You need persistence. You need artifacts that know about you and live under your possession. As a human you need to own your truth. The simplest best artifact that can hold truth about you is files. Files are great, it's essence of everything. They are simple, they are universal, they are great.

Revelation 1: If you want your agent to deeply understand you, you need persistence.

So I decided to give Hermes a shot. And first and foremost I think of it as your personal filesystem which you run an LLM on top of. That's the essence of it. Openclaw is not much different on a broader sense. Just make sure to name it. The way u name your ship etc etc. I named it Bars, it's a snow leopard. It might've been the best decision ever.

What I did before surgery:

Uploaded notes from all my prior sessions with number of surgeons, physiotherapists, nurses.
Store all my perplexity deep research reports on latest breakthroughs and ACL reconstruction practice
Upload all photos
Upload my scans
All information about how the injury happened

It made it easy to meet new specialists and tell them all about my case upfront and in a neat clear form. Every session with a specialist or any new piece of knowledge I gain through research would end up further honing understanding of my case. And as a bonus its readily available to worlds best smartest models in form of files.

After surgery

It's done. It went really well. The medical team at the clinic were incredibly helpful and nice. I was asking hundreds of questions questions per minute.

Right after surgery you get hit with two things:

need to remember to take meds in a timely manner
need to do exercises 5 times a day

Solution was easy: ask Bars to create a number of cron jobs that remind me on meds and reminds me on exercises. Amazing.

After one gets discharged from the hospital there is a weird paranoia about everything that happens to your leg. And there is always tension between phone/email ur clinic asking about every detail vs just closing your eyes for some small things.

I decided to stay paranoid for best of my knee. I asked my agent every little thing. I think models right now are very good at not having a strong opinion when it comes to things like healthcare. It won't tell u a single answer but it will tell you to take a note of things, keep log of how you feel, take photos every night, get all surgery details in order.

I set another cron that would ask me about my pain levels and tell me to take a photo of my knee each day.

I realised being mostly on bed means I'm not moving much so I decided to start tracking my calories to take care of my diet and help my body with as much variety of resources as possible. I kept forgetting to log calories. Solution → another cron job that asks me to log calories every breakfast, lunch dinner.

I decided to go all in with monitoring my health in general and connected my Oura to Bars so that it sends a sleep report every morning.

Telegram channel is pretty great but still not mouldable interface. I wanted to actually see and feel the data so I ended up creating a personal health dashboard. It's like Grafana but for health. It now has all historical data (90 days worth) on my weight, calories, exercises, knee swelling/extension visialized. The best thing is that it is only accessible to me through Tailscale.

Now after I set dozens of notifications up I ran into maybe the biggest problem so far. I ran out of attention. I got overwhelmed with notifications that I naturally just ignored it.

I knew what I needed to do. I knew when I needed to do it. I had the recipe, the most optimal policy. But I just didn't do it cuz of basic fatigue.

I didn't want to go to AI anymore. Now I needed the AI come to me.

Small sneak peak into it

Talk more about it in next part….