Context

This article expands on a reflection published on LinkedIn on March 10, 2026. It reflects my thinking at that time. Since then, my practices around AI agents, their memory, and their identity have evolved. This experience remains the starting point of that evolution.

The LLM, a Mirror of Our Identity

When I left ChatGPT for Claude, I thought I was simply switching tools.

I expected to find the same experience: an assistant able to answer my questions, challenge my ideas, and help me structure my thinking. Like replacing one code editor or one browser with another.

Yet a few days later, an unexpected feeling set in.

I hadn't just switched LLMs.

I had lost something I'd never been aware of building.

That experience led me to a question that still stays with me today:

At what point does an LLM stop being a simple tool and become a reasoning environment?

A transition I thought would be seamless

I didn't leave ChatGPT to compare models or run an experiment.

At the time, I simply thought Claude would fit my needs better. The decision was pragmatic.

I wasn't looking for a new way of working.

I wanted the same relationship, with a different assistant.

I thought a few hours would be enough to pick my habits back up.

That wasn't the case.

What actually surprised me

The first conversations were relevant.

The answers were good.

The capabilities were clearly there.

And yet, something felt off.

I kept coming back to Claude with the feeling that I had to rebuild a way of exchanging ideas that, until then, had become natural to me.

Of course, I knew a new assistant wouldn't have my history.

But that wasn't what was bothering me.

What I was missing was harder to describe.

I couldn't find the fluidity I had gradually built over hundreds of exchanges.

My way of asking a question.

Of introducing context.

Of developing an idea.

Of coming back to a line of reasoning several days later.

I was discovering that this way of working hadn't been built in my head alone.

It had been built in the relationship itself.

That's when I understood I hadn't just switched assistants.

I had left behind a reasoning environment that had become familiar.