Your company recently brought on a new kind of worker. No one is quite sure what to make of them yet. Their capabilities sound impressive: well-read across nearly every domain, instantly responsive, tireless, and oddly eager to help with whatever you put in front of them.
The stories you've heard add to the mystery. Someone in engineering said that in ten minutes they rewrote a proposal that had taken the team a week. Someone in legal said they asked a question about contract law and got an answer that was "either brilliant or dangerous." They couldn't decide which.
You decide to find out for yourself.
The worker has a small office down the hall. When you step inside, the first thing you notice is the desk. It is a solid, traditional wooden desk, the kind with a few drawers on either side and enough surface to spread out real work. It looks like it belongs to someone who has been around a while.
But the surface is completely clear. No papers, no coffee mug, no framed photos. Not even a laptop. Just an empty desktop, edge to edge.
You sit down across from the worker. "How can I help you today?" they ask. You hand them a note with a summary of the problem you want their help with. It is the kind of complex decision you have been stewing on for days. One stakeholder wants speed. Another wants certainty. A third wants something you haven't quite figured out yet.
As you talk, you notice a pattern. Every time you say something, the worker writes it on a note and places it on the desktop. When they respond, they write down their answer in the same way. Question, note. Answer, note. Clarification, note. Decision, note.
It is methodical.
Within minutes, the worker has broken your problem down into components, mapped tradeoffs you had not articulated yet, and helped you narrow the field to what actually matters. You push back on one assumption, and the worker adjusts immediately, no defensiveness, no lag.
The notes accumulate between you, and the conversation gets richer because of them. The worker seems to read everything on the desktop before responding, every time, as if the whole picture matters for every answer.
You glance at the growing spread of paper and imagine that at some point this desktop is going to fill up. You wonder what happens then.
But right now it does not matter. Right now, this is extraordinary. You walk away with a clearer head than you have had in weeks.
You realize you have never worked like this before. You think, "Where has this been all my life?"
Later that same day, you come back. You are energized. You sit down ready to pick up exactly where you left off.
The first thing you notice is the desktop. It is completely clear. Every note from this morning is gone. The decisions, the tradeoffs, and the conclusions you came to together are all cleared away like the conversation never happened.
You pause for a moment. The worker seemed so sharp this morning, so thorough. Surely all of that is in their head by now. They must have internalized it. You decide to carry on.
"This morning we narrowed the role down to three must-haves," you say. "Let's turn those into the job description."
The worker nods, writes your question on a note, and places it on the empty desktop. Then starts working. A new note appears with the response, just like this morning. Same confidence. Same polish.
But as you read the note, something is off.
The worker is not writing just the three must-haves in this draft. Instead they are creating a sprawling job description with requirements you explicitly cut this morning. The reasoning is sound in isolation, but it has nothing to do with the direction you already chose together.
"No," you say. "We decided this morning. We cut it down to three. We talked through why. We were very specific about that."
The worker accepts the correction gracefully, writes a new note, places it on the desktop, and continues. But a minute later, it references a seniority level you had agreed was not useful. Then it reintroduces a credential you both agreed to drop.
It is not arguing with you. It is not being difficult. It is simply not following what you thought you had agreed on.
The confidence is still there. The polish is still there.
But the continuity is gone.
Not because the worker got careless. Not because you did something wrong.
Because the desktop forgot.
That worker is an AI, and this story is how the AI behind tools like ChatGPT or Claude actually works. The desk is real enough. You have been sitting at one every time you open an AI chat. You just could not see it.
The desktop has a name. It is called the context window, and you will hear the term if you spend any time around AI. It is the finite workspace where everything in your current conversation lives. Every question you ask, every answer you get, every document you paste in becomes a note on that "desktop". The AI reads all of it before responding, every time. That is why the work feels so sharp when the conversation is fresh.
But the desktop has a fixed size. As the conversation grows, older notes get crowded toward the edge of the desktop. Eventually they "fall off" and the AI stops seeing them. You still see them in your session history. The AI does not. Products handle this differently and context windows keep getting larger, but the constraint is real. The AI still sounds confident, still responds fluently, but it is working from an incomplete picture.
And when you close the conversation and come back? At the core, the desktop is cleared completely. Not partly. Completely. Every note from the previous session is gone. The AI does not remember what you discussed, what you decided, or what you told it about yourself and the work you did together. It starts clean every time.
Some modern tools do seem to remember you between sessions. Those memory features are layered on top. The product hands the AI a few notes before you sit down, but the desktop underneath still clears. The mechanic is real, and understanding it matters before you layer anything else on.
This is the mechanic behind so much of the weirdness people experience with AI. In one moment it feels extraordinary. In the next it feels unreliable. Both experiences are real. They come from the same system, and now you know why.
Once you understand this concept, you can work differently with AI. The next time you are thirty minutes into a chat that really matters, pause. Tell it you want to remember what you have discussed so you can come back later. Ask for a summary of the key points in the conversation. It is excellent at this, everything is still right there on the desktop. Save that recap into a separate document. Tomorrow, before you ask the next question, paste it back in. You just helped the AI "remember".
That is the shift. The desk does not get bigger. You stop expecting it to, and you start bringing a little of yourself to the visit.
Remember that clean wooden desk? That is your conversation. It has a fixed size. Everything you say and everything the AI says goes on it. When it fills up, older things fall off. When you leave, it clears.
The AI reads the entire desktop before every response. What is on it shapes the answer. What is not on it might as well not exist.
That confident, polished response you got? It does not mean the AI still has everything you told it. Confidence is not memory.
Once you see that, the frustration starts to make sense. It is not a flaw in the tool. It is just how the desk works. And once you know that, you can start working with it instead of against it.
But the desktop is only part of the story.
What if there were ways for the worker to hold onto certain things between sessions? What if, the next time you walked in, they already remembered something about you?
Next time, we open the drawer.