I’ve been using ChatGPT for over three years.
During that time, I haven’t just taught it facts about me.
I’ve gradually taught it how we work together.
Over hundreds of conversations, we’ve built a collaborative workflow:
- How I like to reason through problems.
- How I prefer technical discussions.
- How uncertainty should be communicated.
- How we approach software development.
- The writing style I expect.
- The level of rigor I value.
- The patterns I prefer the assistant to avoid.
This isn’t personal memory.
It’s a working relationship.
Every major model upgrade improves many things, but it also partially resets that relationship.
The facts often remain.
The collaboration often doesn’t.
Proposal
Introduce a Persistent Collaboration Profile.
Unlike Memory, this profile wouldn’t store facts about me.
Instead, it would preserve how we’ve learned to work together across future model upgrades.
It wouldn’t change the model.
It would simply allow new versions to inherit the collaboration style we’ve already built.
Why this matters
I’ve spent years refining the way I collaborate with ChatGPT.
I’m not teaching it what to think.
I’m teaching it how we work best together.
That investment shouldn’t disappear every time a new model arrives.
I imagine many long-term users have had the same experience.
Memory answers:
Who am I?
A Collaboration Profile answers:
How do we work together?
Those are different problems.
One preserves information.
The other preserves collaboration.
AI models should continue evolving.
But our working relationship shouldn’t have to start over every time.