Volume 1
How to turn your computer into a working environment you can control in plain language
You have used a computer for twenty years. Maybe thirty. You have written dissertations, invoices and apology emails. But no one has ever explained what is really inside that box.
This book does not teach you to program, and it does not teach you to be a prompt wizard either. It teaches you to stay in control: while the artificial intelligence opens folders, moves files and writes lines of code, you are already thinking about the next step, or about how to make the next time take half as long.
The part that used to be hard, it does. The part that is still hard is yours: saying whether what it did is right.
Why this book exists
AI can use a computer far better than you can, and that is exactly why learning to talk to it is worth the trouble. It does not need to get better: you need to get more precise.
For thirty years the software industry worked to make the computer easy, and making it easy meant taking it out of sight: icons, windows, round reassuring buttons. It was the right call, and it held up perfectly well while you were the only one using the machine.
But that simplification was made for eyes, and AI has no eyes on your screen. It has no hand on the mouse, it cannot double-click the icon, it has no idea what "that file on the desktop, the blue one" means. It works on the floor below — the one they hid from you for thirty years — where everything has a name and an address.
The computer has become a workspace for two, and it works to the extent that both can understand it. The machine is already down there: the one who has to take a step towards the other is you. This book is about that step, and about getting you to look at the machine this way until it stops feeling strange.
Everything else follows from there: if you know how the place where the two of you work is made, you also know what to ask, what to expect, and when to stop the machine.
The point
On a subject where you are competent, this thing is an extraordinary interlocutor, because you can weigh its objections. On a subject where you know nothing, the conversation looks exactly the same: same fluency, same confidence.
For the first time the power of these tools is not reserved for the people who build them. Things that ten years ago required an IT department can now be asked for by anyone, in their own words. That is a real liberation, and it should not be played down.
Except that until yesterday doing and judging came together. Someone who could do a difficult thing could also say whether it had turned out well: it was the same competence, you learned to judge while you learned to do. Now the two have come apart. The doing you can get from somewhere else. The judging stays where it was.
And you use the machine mostly where you fall short: you did not get it for the work you already knew how to do alone. But outside your field you have nothing to hold on to when deciding whether it is right, and it never sounds any different.
A misspelled name is visible. A wrong number is not: it looks just like a right one. And a wrong judgement has the air of a right one.
The step this book takes you up is that one, and it is not where you expect it: it is not there to make you produce more, which is by now the easy part. It is there to put you in a position to check. That is where the four skills below come from — they are not tricks to memorise, they are what it takes to stand in that position.
What you learn
Learning to work with a computer that can understand instructions, use tools and act alongside you. There are no formulas to copy: there are gestures that stay valid when the tool changes, and the book has you practise them one at a time.
01
Folders, files, paths. It sounds like bureaucracy and it is the most important notion in the book: a machine that cannot see your screen has no idea what "that file there, the blue one" means. By the end you can point anything out, even out loud.
02
The terminal is the kitchen the AI cooks in. Someone who has never been inside only sees the plate coming out, and at that point they either trust too much or not at all. You learn to stand in the doorway: what it is doing now, where it is putting its hands, when it is worth stopping it.
03
Describing a result instead of a task, giving the right references, noticing when your request can be read two ways. It is the part that stays yours even when the work moves over to the machine.
04
Telling a good result from one that merely looks good: counting, looking at the extremes, checking the order of magnitude. And being able to go back to when it worked, knowing which few things cannot be undone.
Annotated contents
Volume 1 is complete. It can all be read here, with no sign-up and without leaving an email address.
Store, transform, communicate: everything a machine has ever done fits in three verbs. Then the archive — where your things actually ended up — and the path: a file's address, which is how you will point things out to an AI.
The terminal is not dangerous in itself: it is about as dangerous as your kitchen. You go in, you look around, you move between folders. In the whole chapter there are only two commands capable of doing damage, and they sit inside two thick-bordered boxes.
The first half explains what these tools materially are, what they see and how they fail: that is the mental model that lasts. The second half names names, tells the products apart and compares costs: that is the part to consult and keep updated.
The first time you watch it work it makes an impression: the wonder and the knot in your stomach are equally wrong. They have the same cause — not knowing what is going on. It has not become smarter: it has been given hands.
How it is built
Every concept closes with three lines: the concept, a short task, and an observable test you can apply to yourself. That is the difference between reading a book and learning something.
The exercises are done at the computer, not in your head. That is why the site is built to sit in half a screen, next to a terminal window: nothing breaks below five hundred pixels, nothing is hidden, nothing asks you to sign up to keep reading a page that is already open.
$ pwd
$ lsTwo commands, and you know where you are and what is there. The book starts here and never takes the previous step for granted.
Who it is for
There is no professional profile this book is aimed at. There is an attitude: you use a computer to do things you care about, and you suspect there is far more inside that box than you have been shown. The terminal strikes you as obscure and powerful. Obscure it is not. Powerful it is — and that is exactly the reason to learn your way around it.
Frequently asked
They are here in plain sight, not inside a panel you have to open: if you need an answer, you find it by searching the page.
From the blog
The eight chapters can already be read online in full: the blog is the material around them, made of real cases, procedures to redo on your own work, and starting points. Below, the latest piece from each of the three strands: a recipe to follow from start to finish, something to understand before getting your hands in, and a tool to connect to your agent.
The articles are in Italian. The commands, the file names and the tools in them are not: if that is what you came for, they are usable as they are.
Come dare a un agente da terminale l'accesso in lettura alle query, alle impressioni e alle posizioni del tuo sito, sapendo che qui un server ufficiale non esiste e cosa comporta.
Sono il file che spieghi una volta sola all'agente, così non lo rispieghi a ogni sessione. Cosa ci va dentro, cosa non ci va, perché i nomi sono due e come scegliere quale usare nel tuo progetto.
Exa, Tavily, Perplexity e la ricerca web delle API Anthropic e OpenAI: cosa cambia davvero, quanto costano con i numeri di una prova vera, e come capire se ti servono prima di collegarne uno.
Free ebook
Usually what you download in exchange for an email is an extract, or a taster chapter. Here it is the whole of volume 1, in EPUB and PDF, with every chapter and figure: the same text you read on this site, to carry on a reader, consult offline and annotate however you like.
You also get an email when we publish new chapters, practical recipes or significant updates to the project. We write only when there is something worth reading: no constant promotional email, and your address is never passed on to anyone.
In preparation. The ebook is not ready yet: subscribe now and you get it as soon as it is out, with nothing else to do.
The ebook and the emails are in Italian. The web edition of the book is available in English; subscribing is worth it if you read Italian or are happy to run the messages through a translator.
Rather not leave your email? The translated chapters are also in the English RSS feed. The ebook is not: that one goes by post.
