The Model Has No Memory: It Starts Over Every Time

The collaborator who leaves the room

Let's go back to our collaborator, with one more rule that sounds cruel: at every single move it steps out of the room and comes back remembering nothing. Not yesterday: thirty seconds ago.

Did it just read a file? Steps out and back in, doesn't remember. Did it just run a command and see the error? Doesn't remember. Did you just explain what you want? Doesn't remember.

So how is it possible that everything fit together the way you saw in the last lesson? Because every time it comes back in, someone redoes the recap from scratch: who you are, what you asked it, what it's already done so far, how each move went. Every time. All of it.

That continuity you see scrolling across the screen isn't the model's memory: it's a recap rebuilt from zero every turn.

What it seems like and what it is

It's the most common wrong belief, and it's worth naming to get it out of the way:

• What it seems like: "While we work together it learns how I work, and the further we go the better it knows me."

• What it actually is: it learns nothing and doesn't know you. It only knows what gets replayed to it in that moment. If it seems like it knows you, it's because someone is reminding it every time.

What this changes for you, in practice

This isn't a technical curiosity: it's the root of three things you'll run into constantly.

• If you close and reopen, you're back at square one. New session, blank page. The things you don't want to repeat every time need to be written down somewhere — there's a chapter just for that.

• The longer the conversation, the more there is to replay. And if there's a lot of stuff in there that's no longer needed, the important things carry less weight. That's why, past a certain point, it's worth cleaning up instead of piling on.

• Whatever you haven't told it doesn't exist. It hasn't "picked up on" how your work is organized: either you tell it, or it's as if it isn't there.

And now

One question remains, and it's the big one: who is it that redoes the recap every time? Who hands it the tools, who keeps track of the conversation, who actually runs the commands on your computer. It has a name, and we'll see it in the next lesson — because it's that thing, not the model, that's the real reason Claude Code works so well.

Did you get it?

Two minutes, no grade. Trying to answer — even getting it wrong — cements the concepts far more than rereading. Not in the mood? Skip it: the lesson still counts as complete.

1Does the model remember what you did two moves ago?

Yes: it holds everything in mind during a session No: it starts from zero every time, and the conversation gets replayed to it from scratch Only if you explicitly ask it to

2What happens when you close Claude Code and reopen it the next day?

It picks up where you left off, like a chat It starts fresh: it knows nothing of what you told each other yesterday, unless you left it written down

3Why does a very long conversation make answers worse?

Because the model gets tired and loses precision Because every move it has to reread everything, and the more useless stuff is in there, the less weight the things that matter carry Because the session expires after a certain number of messages