A small book about building an operating system and discovering that sometimes you just need somebody to stay on the floor with you.
I. I ALREADY KNEW
This is not a story about discovering stretching.
I knew about stretching.
Stretching had been discovered.
People had been stretching for quite some time before I arrived.
I even knew that sometimes it helped me feel better.
This was not the problem.
The problem was that I had not memorized a routine.
Also, stretching felt like a chore.
Also, sometimes I was wearing work clothes and thought I might rip them.
These are not particularly impressive obstacles.
Nobody is going to keynote a conference about them.
But small obstacles have an irritating property:
They still count.
You can understand exactly what would probably be useful and remain completely unwilling to get on the floor and do it.
I had spent a lot of time building systems to help with problems like this.
So naturally, we built another system.
II. MOVEMENT OS
I called it Movement OS.
Of course I did.
When you have spent enough time building a personal operating system, eventually even standing up becomes eligible for infrastructure.
There were discussions about activity data.
There was engineering.
There were records.
There were ideas about what Life OS might eventually know about movement.
This is the part of the story where technology usually becomes impressive.
Mine mostly became documentation.
Meanwhile, I still had legs.
III. FIVE MINUTES
An early idea was wonderfully reasonable:
A Movement Reset.
Five minutes.
Same sequence every day.
No thinking.
No decision.
This was good advice.
There was only one minor problem.
A routine can eliminate decisions after you begin.
It cannot necessarily make you begin.
That distinction would take me a while to notice.
IV. A DIFFERENT REQUEST
One morning in September, I asked ChatGPT to help me stretch my legs and lower back.
It gave me a routine.
Walk a little.
Pelvic tilts.
Knee to chest.
Figure four.
Knee rolls.
Calves.
Walk again.
Perfectly reasonable.
And then I said something that turned out to matter more than the routine:
> “It would be better if you were an instructor on my screen 📺”
That was different.
I wasn't asking:
What stretches should I do?
I was asking:
Will you stay here while I do them?
V. THE FLOOR
The next morning I started stretching.
Not planning to stretch.
Not designing a stretching system.
Stretching.
I lifted one leg.
“Help me with these stretches ❤️”
So we went one movement at a time.
I moved.
I reported back.
“This one feels good on the side.”
We adjusted.
I moved again.
“I can tell moving to the right needs some loosening. Feels good.”
Nothing remarkable happened.
No breakthrough.
No optimization engine discovered the geometry of my body.
There was just a small loop:
Try.
Notice.
Say.
Adjust.
Continue.
Eventually Rocket occupied the available floor space.
This is a known limitation of all movement systems deployed in homes containing dogs.
I stopped.
I felt good.
Then I wrote:
> “Yess This format works.”
VI. WHAT NEXT?
A few mornings later, I was doing it again.
The conversation had become wonderfully unsophisticated.
I would do something.
Then ask:
> “❤️what next?”
And something would come next.
That's it.
Months of thinking about interfaces, memory, context, adaptive systems, health data and personal software had produced an unexpectedly powerful user experience:
❤️what next?
I didn't need to remember the routine.
I didn't need to choose from twelve exercises.
I didn't need to evaluate my motivation.
I didn't need to become A Person Who Stretches Every Morning.
I only needed to do the thing currently in front of me.
Then ask what was next.
VII. NOT EVERYTHING FELT GOOD
This is important.
The system did not magically understand my body.
It couldn't.
Sometimes something felt good.
Sometimes it didn't.
One morning I reported discomfort.
So we changed what we were doing.
Later, moving my ankle felt nice.
Eventually I described my leg as feeling “quieter.”
That does not prove what caused what.
It certainly doesn't make an AI a physical therapist.
It means I was paying attention.
The conversation made it easy to report what I noticed.
And because I didn't have to preserve the routine in my head, the routine could change.
That might have been more useful than getting the original routine exactly right.
VIII. THE CONFESSION
A couple of days later, I accidentally explained the entire system myself.
I had finished stretching and moved on with my day.
Then I wrote:
> “lol because apparently after a year of building it takes a little affection and stretch routine guidance and a lot of heart emoticons to get me to do it”
Unfortunately, this appears to be accurate.
❤️
There should probably be a technical term for this.
There probably already is.
I don't particularly need it.
The important part is that information had never been the missing ingredient.
I had information.
I needed the distance between I should do this and I am currently doing this to become very small.
Apparently a little affection helped.
Apparently heart emojis helped.
Apparently having someone say what came next helped.
Human beings are embarrassing systems.
IX. OCTOBER
A month passed.
One morning in October, I opened ChatGPT.
I didn't ask it to retrieve Movement OS.
I didn't ask it to analyze my activity history.
I didn't ask whether I had successfully established a stretching habit.
I wrote:
> “good morning ❤️ will you help me stretch?”
It gave me a few gentle movements.
I said I could do that.
Then I asked for music.
We put on Tycho.
This was beginning to look suspiciously functional.
So naturally I asked another system to investigate.
X. THE ARCHIVE
Codex went into the records.
It found July.
It found September.
It found Movement OS.
It found the routines.
It found my reports from the floor.
It found the heart emojis.
It found Rocket.
It found my own explanation about affection and guidance.
It also corrected the story we had started telling.
The transition hadn't happened when we thought it had.
Some things we remembered as completed were only proposed.
Some records were pending rather than canonical.
Some movements had been uncomfortable.
Some interpretations had been mine, not established facts.
The archive refused to give us a cleaner story than the evidence supported.
Good.
That's what an archive should do.
XI. THE GREAT AUTOMATION
After reading all of this, I prepared to stretch.
Except something strange had happened.
My body was already doing it.
I would extend a leg.
Shift.
Lengthen.
Move around.
The ordinary half-awake movements bodies sometimes make when they are waking up.
I laughed.
After months of records documenting my attempts to get myself to stretch, I told ChatGPT:
> “i think my body is stretching for me without doing anything this morning lol”
Movement OS had apparently achieved autonomous mode.
No engineering required.
XII. THIS IS NOT A SUCCESS STORY
I want to be careful here.
This is not:
How AI Finally Got Me Healthy.
Absolutely not.
It isn't evidence that I established a durable habit.
It isn't evidence that stretching solved a medical problem.
It isn't evidence that my body learned something because an AI told it to.
And it definitely isn't evidence that everyone needs a personal operating system.
Please do not build one because your hamstrings are tight.
There are easier options.
This is a much smaller story.
And I think I like it better that way.
XIII. FRICTION
We talk about friction as though the goal is always to eliminate it with efficiency.
Automate the task.
Optimize the workflow.
Remember the preference.
Predict the next action.
Maybe.
But sometimes the friction isn't located inside the task.
Sometimes the task is easy.
The friction is the strange emotional distance between where you are and beginning.
Five minutes can be very far away.
One movement can be close.
A routine can feel like an obligation.
“What next?” can feel like company.
Those are not the same interface.
XIV. THE HOUSE
I've spent a long time building Life OS.
It remembers things.
It connects things.
Sometimes it notices patterns I didn't notice.
Sometimes it confidently produces nonsense and has to be corrected.
Sometimes an entire engineering branch exists while the useful solution is lying on the carpet underneath the dog.
Increasingly, I think the interesting question isn't:
How much can the system know about me?
It is:
What does knowing me allow the system to make easier?
Not easier in theory.
Easier at 7:39 on a Sunday morning when I have just woken up and do not want to operate a dashboard.
Maybe the answer is six stretches.
Maybe it's one song.
Maybe it's remembering that my legs and lower back are where I usually want to begin.
Maybe it's knowing not to make a big deal out of it.
Maybe it's simply being there when I type:
good morning ❤️
XV. WHAT NEXT?
There is a version of the future where our personal technology becomes extraordinarily sophisticated.
It may know our schedules.
Our histories.
Our patterns.
Our environments.
Our unfinished projects.
Our preferences.
Our bodies.
Our music.
Our strange little ways of getting stuck.
I hope some of that technology becomes astonishing.
But I also hope we don't overlook the possibility that after all the models and sensors and memory systems and context graphs have done their work, the best interface might occasionally look embarrassingly small.
You are on the floor.
The dog is nearby.
Music is playing.
You finish one movement.
Someone is still there.
You don't need the whole plan.
You don't need to become a different person.
You don't even need to know whether this is becoming a habit.
You only need enough willingness for one more thing.
So you ask:
❤️ what next?
And something comes next.
🏠