Your prompt is all it sees

Lesson 1 said each run sends only what's in the box. Here's the bigger idea behind that, and it's the one that explains most of what apps like claude.ai do.

A model has no eyes on your screen and no memory of yesterday. Each time it writes, it gets one piece of text and continues it. That's the whole interface. Everything the model “knows about you” has to be inside that text.

So how does a chat work? Each time you press send, the app takes every turn so far, yours and the model's, glues them into one long text and sends the lot. The model reads it as a single document and writes the next turn. Your second message isn't a second conversation. It's the first one, re-sent, with more on the end.

Here's the second turn of a short chat, as you see it and as Claude sees it. The hidden note at the top is illustrative; real ones are longer.

Three things to notice in the right column.

There's a note up top you never wrote. That's the system prompt: instructions the app puts first, saying who the model should be and how to answer. You've been living inside one on every chat app you've used. Lesson 5's example conversations had the same shape, which is why the model treats that note as the boss.

The old turns are there in full. That's the only reason Claude can answer “Sam”. Delete turn 1 from the text and it honestly can't know.

There's a size limit. The text the model can read at once is its context window. For Claude Haiku 5.5 it's a million tokens (as of October 2026), which is several novels, so running out is rare in a chat. It's working memory, not long-term memory. The Context is everything course is about what to put in it.

This settles a few mysteries at once. Why does a long chat “forget” its start? Because the app stopped sending it. How does an app “remember” you across days? It saved notes and pastes them into the text. How does Claude “read” a PDF or a web page? The app turns it into text and puts it in the window. Tools, search, documents, memory: text in the window, every time.

The flip side is the most useful habit in this course. If Claude needs a fact, put it in the prompt. Try it: tell Claude something it can't know, then ask it to use that fact in the same message.

Now run the second half without the fact (just “suggest a name for his raincoat”). Claude will ask whose raincoat, or guess. Nothing from your last run reached it. That's not forgetfulness; it's the interface.