We have travelled a long way together. When you arrived at the first article in this series, words like "agent", "RAG", and "MCP" probably sounded like a foreign language. By now, you have met the whole cast. Before we say goodbye, let us gather them all in one place — and reveal the one trick that makes every future buzzword easy.
The cast, in one quick paragraph
There is a Librarian sitting behind a wall, who has read everything ever written and can only do one thing: finish whatever is on your note. That note is a prompt — one note in, one note out, no memory afterwards. When you want the Librarian to do something they cannot do alone (check today's news, read a file, send an email), you hire a Runner, which techies call an agent. The Runner stands at the slot for you and handles all the legwork. Stapling background information onto your note is called context. Pasting in records of past conversations is called memory. When the Runner fetches the most relevant document by meaning rather than by keyword, that is RAG (Retrieval-Augmented Generation — "augmented generation" just means the Librarian gets extra material to work with). When the Runner steps outside to ask the internet, that is web search. A fill-in-the-blank form that helps the Runner read the Librarian's requests clearly is called function calling. The standard handshake that lets the Runner ask any tool-shop what it can do is called MCP (Model Context Protocol). A fixed recipe of steps where only the clever bits go through the slot is a workflow. A standing instruction sheet with ready-made scripts for particular jobs is a skill. And a junior Runner sent on a sub-errand is a subagent. That is the whole cast.
The unifying trick
Here is what is genuinely exciting: every single one of those things is doing one of just two moves.
Move A: stuff a more useful note through the slot. Context, memory, RAG, web search, function calling — these are all ways of making the note richer, more specific, better informed. The Librarian can only work with what is written on the note. If the note is thin, the reply is thin. If the note arrives with exactly the right background already attached, the reply is sharp and reliable.
Move B: handle the legwork so you bother the Librarian less often. Agents, workflows, skills, subagents — these are all ways of packaging up the fixed, mechanical steps so the Librarian only gets involved at the moments that genuinely need their brilliance. The Runner checks the timetable, reformats the spreadsheet, sends the confirmation email. None of those steps need a genius. The Librarian only gets called in when you truly need an intelligent reply.
That is it. Two moves. Every term in the AI world, past and present, is either making the note better (Move A) or reducing how often the Librarian is needed (Move B). Many clever-sounding innovations are both at once.
So whenever a scary new term turns up — and they will keep turning up — ask yourself: "Is this making the note better, or is this handling work the Librarian does not need to be involved in?" That question almost always cracks it within a few seconds.
The key line, one last time
The most important sentence in this whole series is one we introduced in the agent article:
The agent is all the parts that do not need a genius.
This is not a put-down of agents. It is a very precise description of what makes them valuable. The Librarian's time (and, as we will see in a moment, their cost) is precious. Every step you can hand to the Runner is a step done faster, more cheaply, and without tiring anyone out. The Runner multiplies the Librarian's usefulness enormously by clearing the decks of everything routine.
When you hear about a new AI tool, the useful question is not "is this smart?" The useful question is "which parts of this need the Librarian, and which parts are just the Runner keeping the desk tidy?"
What it is all for
None of this is technology for its own sake. The whole point is simple: save your time and lower the barrier so anyone can use it.
A well-built AI system should feel, from the outside, like having a knowledgeable friend who never gets tired, never makes you feel embarrassed for asking, and is available at ten o'clock on a Sunday night when your maths homework is due Monday morning. The Librarian, the Runner, the note, the slot — all of that machinery is invisible. What you feel is just the help.
That is the goal. Every piece of jargon in this series is in service of that goal.
A hopeful look ahead
Here is some genuinely good news about where things are going.
Every time the Librarian reads a word, techies call that unit a token. A token is roughly three-quarters of an English word. Tokens cost money — each one takes electricity and computing power to process. Right now, a single note and reply might use a few hundred or a few thousand tokens. That used to be expensive. It is getting cheaper very fast, year on year, in the same way that phone calls used to cost pounds per minute and now cost almost nothing.
There is a real possibility that within a few years, a capable AI model will run on an ordinary home computer rather than a vast data centre. That would change things considerably — no internet required, nothing sent to a distant server, just the Librarian living quietly on your own machine.
And for families who do not want to configure anything at all, pre-packaged tools are already arriving — what you might call super-assistants: a maths tutor, a reading helper, a revision companion, all set up by someone else and ready to use without the family touching a single setting. AI Tutors is built exactly on this idea — the whole system of Librarian, Runner, notes, and slot, packaged so that a twelve-year-old can just open it and ask a question.
You are ready
Here is what you can do now that you could not do when you arrived at the first article.
When a headline says "new AI agent beats human experts" — you know that means a Runner handled the mechanical steps and the Librarian handled the clever bits, and the combination outperformed what humans do alone. When someone mentions "RAG-based search" — you know the Runner is fetching relevant documents and stapling them to the note so the Librarian has better material. When a product brags about "workflow automation" — you know it means the fixed steps are baked in, and only the moments that need intelligence go to the Librarian.
None of it is magic. None of it is mysterious. It is a very sophisticated Librarian, a diligent Runner, and an ever-improving system for getting useful notes through the slot.
You now understand the whole picture better than most adults who use these tools every day. Whatever buzzword appears next year — and there will be one, because this field moves fast — you have the question ready: "Is this a better note, or less bother for the Librarian?"
It will almost always be one of those two things. Probably both.
This is the finale of the AI in Plain English series. If you are arriving here for the first time, the best place to start is AI Is Just a Word-Guessing Machine — the very first article, which introduced the Librarian and the slot. Everything else builds from there.