Imagine your local library has one very clever Librarian. They sit behind a wall. The only way to reach them is through a mail slot — you slip a note in, a reply slides back out. No eye contact, no small talk, one note at a time.
Now imagine that same library decides to open three different doors on three different streets. One door has a friendly counter where anyone can walk up and write a note. Another door leads to a back office used mostly by builders and engineers. A third door opens into a room where a helper-on-call can also rummage through your filing cabinet for you.
Same Librarian behind the wall every time. Completely different front-door experience.
That is exactly what is happening when you hear people talk about AI chat tools, coding assistants, and desktop AI helpers. The brain — the model (what techies call an LLM, or Large Language Model, which just means a very large word-prediction engine trained on enormous amounts of text) — is the same kind of thing in every case. What changes is the window you are standing at.
Window one: the chat box
The chat box is the simplest window. You open a website or an app, you type something, and a reply appears. No installation, no fuss. This is what most people encounter first.
Under the bonnet, someone has built a neat front counter. You type your note. The counter passes it to the Runner — the agent (a piece of software that handles the mechanical legwork) — who stamps it with any useful background information and slips it through the slot to the Librarian. The reply comes back, the agent tidies it up, and the counter shows it to you.
The Librarian does not know you came in through the chat window versus any other door. They just see a note.
What the chat box adds on top is mostly comfort: a scrolling history so it looks like a conversation (though as we covered in the memory article, that history is just being re-pasted onto every new note), a clean text box, sometimes the ability to upload a picture or a document. The window makes it feel warm and conversational. The Librarian is still doing one note at a time.
Window two: the coding tool
A coding tool — sometimes called a coding assistant — lives inside the software a programmer uses to write code. It is the same library, but the door opens directly into the workshop.
Why did this window appear for programmers first? Because programmers were among the earliest people to realise that the Librarian is extraordinarily good at pattern-matching text, and code is just very precise text. Ask the Librarian to "continue this half-written function" and they are surprisingly capable, because they have read millions of examples of similar code.
The Runner here does a bit more than in a basic chat box. When you are halfway through a function and you ask for a suggestion, the Runner quietly gathers the relevant files you have open — a bit of context you did not have to type yourself — and includes them on the note before it goes through the slot. The Librarian sees more of the picture and the reply is more useful.
There is a gentle lesson hiding here for non-programmers: the coding tool is not magic, and the Librarian inside it is not a different, smarter Librarian. They made the same mistakes with code as they do with prose. They can suggest plausible-sounding nonsense just as cheerfully as they suggest the right answer. The programmer still has to check.
Over time, the people who built coding tools noticed that "help me write a line of code" and "help me draft an email" are not as different as they sound. Both are just notes through the slot. And so many coding tools quietly grew into more general assistants — not by swapping the Librarian, but by teaching the Runner to fetch more kinds of things and giving the front door a wider welcome mat.
Window three: the desktop or everyday assistant
The third window is the one that finally feels like what science fiction promised: a helper that sits on your computer, knows about your files, can open your calendar, write a draft, search the web, and generally act on your behalf.
The Librarian is still the same. But the Runner here is much more capable. Remember from the function calling and MCP article how the Runner can visit tool-shops — pieces of software that offer specific services — using a standard handshake? A desktop assistant has a Runner that has been introduced to many more tool-shops: your files, your browser, maybe your calendar or email.
What makes this feel different — what makes it feel like a proper assistant rather than a gadget hidden in a cupboard — is that you can usually see what the Runner is doing. A good assistant shows you which tools it is about to use and asks permission before it digs into your documents. That transparency matters. The Librarian is still behind the wall; the Runner is no longer invisible.
This is also where the idea of "giving the assistant a friendly home" becomes important. An assistant that can see your files but has no clear boundary around what it is allowed to touch feels alarming. An assistant that says "I can see your Downloads folder and your calendar, and here is what I propose to do — shall I?" feels manageable. Same capability; very different experience.
The window is not the brain
Here is the single most important idea in this article: the name of the product is not the name of the AI.
People often say things like "I asked [Product Name] and it told me..." as though the product and the AI are the same thing. They are not. The product is the front door, the lobby, the Runner, and all the design decisions made about what the slot looks like. The AI — the Librarian, the model — is a separate layer underneath.
This matters in two practical ways.
First, two products can use the same underlying model and feel completely different, because one has a basic counter and one has a sophisticated Runner with ten tool-shops. The Librarian has not changed; the building around the slot has.
Second, a product can change its underlying model without changing its name at all. When that happens, the Librarian improves (or occasionally changes personality). Everything you learned about prompts, context, and memory still applies, because those ideas live at the slot, not at the front door.
You never need to code
If you are a parent reading this and feeling vaguely anxious that you might be locked out of the "real" AI because you cannot write code, here is some reassurance: you are not.
The coding window is genuinely useful for people who write code. But the chat window and the desktop assistant window are just as real, just as capable in their own lanes, and require nothing more technical than typing a message. The same ideas from this series — write a clear note, give useful context, remember that the Librarian forgets — apply in every window.
Pick the window that suits the task. Everything else stays the same.
What we have covered
The Librarian behind the slot is the constant. The windows — chat box, coding tool, desktop assistant — are the variables. Some windows give the Runner more tool-shops to visit; some show you more of what the Runner is doing; some are aimed at a particular kind of note (code). None of them change the fundamental nature of the Librarian or the slot.
Once you see this, a lot of the breathless product announcements become easier to decode. "New AI assistant with file access!" just means: Runner with a new tool-shop. "AI built into your editor!" just means: Runner that auto-gathers code context before sending your note. Same library. New doors.
Next up in the series: What Is an AI Workflow? — when the errand is always the same, you fix the steps into a recipe, and only the thinking step needs the Librarian.
Back to the AI in Plain English series hub or home.