Your Working Setup

The surfaces get renamed and repriced. The folder is the part you keep.

1. Name the five surfaces your tool offers and say in one line what each is for.
2. Say which surface holds your standing context, and load a file into it.
3. Decide whether a paid plan would change anything for the work you actually do.
4. Write an about-me.md that a stranger could read and then brief someone on your behalf.
5. Decide whether a particular piece of information should go into a tool, using a question rather than a memorised list.
6. Name one situation where loading your standing context makes the output worse.

ai-kit/about-me.md — a short file covering who you are, what you work on, who your output is for, how you want things written, and what you never want back. Every module after this one reads it.

Most people's first month with AI is spent horizontally: an account here, a trial there, the same question asked five times to see who wins. It feels like research. It teaches you almost nothing, because the thing that separates a useful answer from a useless one at this stage is not which service produced it.

Go vertical instead. Choose one mainstream assistant and use it for everything for a month, including the things it will be bad at, because finding the edges is the point. The depth you get from one tool — knowing where its file upload chokes, knowing what its standing instructions actually change, knowing how it behaves when you push back — is worth more than a shallow acquaintance with four.

Choosing is not a big decision, and it is reversible. Every skill in this course is a skill about briefs, context, checking and standards. None of it is a skill about a menu. When you switch tools next year, and you will, you will carry the folder across and be productive on the new thing in an afternoon.

Whatever you picked, it gives you five ways in. Learn them as categories. The names on the buttons will change; the categories have been stable for years.

Chat. A message box and a running conversation. This is where most work happens and where most people stop. The important property is that the conversation is the material — everything said so far is part of what the next answer continues from, which is why a long messy thread starts producing long messy answers.

File upload. Attach a document and ask about its contents. This is how you convert a question it would have to guess at into a question it can answer from material in front of it. Nearly every tool takes PDFs, spreadsheets and images now. This surface is the single biggest quality upgrade available to a new user, and it is one click away.

Persistent instructions. A place to write things once that apply to every conversation afterwards. Every mainstream tool has this surface under some name of its own; ChatGPT calls it custom instructions. This is where ai-kit/about-me.md goes. It is the difference between explaining your job at the top of every chat and never explaining it again.

Web access. A mode where it can fetch pages rather than answer from what it absorbed during training. Turn it on for anything time-sensitive, and treat what comes back as a set of links to check rather than a verdict. Module 09 is about doing this properly.

Voice. Talking instead of typing, on a phone app, usually while walking or driving. Voice is excellent for thinking out loud, terrible for anything where the exact words matter, and worth knowing about because it converts dead time into drafting time.

Remember: the buttons move. If your tool has renamed persistent instructions or buried file upload behind a menu, look for the category, not the word. Every mainstream product still has all five.

Do the whole of Part 0 on a free account. Nothing in the first four modules needs a subscription, and you should not pay for something before you know whether you will use it.

When you do consider paying, ignore the feature grid and ask what changes for your work. In practice a paid plan buys three things: the better model rather than the fast cheap one, higher limits so you do not get cut off mid-task, and earlier access to the surfaces above. The quality difference between the free model and the paid one is real, and it is smaller than the difference between a two-line request and a proper brief. Get the brief right first. If you are still hitting limits or still dissatisfied after Module 02, that is the signal to pay, and by then you will know what you are buying.

Chat histories are a bad filing system. They are searchable in theory and unsearchable in practice, they belong to a company rather than to you, they cannot be diffed or versioned, and they do not survive you switching tools. Anything you intend to reuse should be a file on your own disk.

Make the folder now.

ai-kit/
  ai-map.md        <- from Module 00
  about-me.md      <- you write it in this module

Plain text and markdown, nothing clever. It opens in any editor, syncs with whatever you already sync with, and can be pasted into any tool that exists next year. Later modules add files beside these two. By Module 11 the folder is the thing you actually own, and every module in between writes into it.

The file answers the questions a competent new colleague would ask on their first morning, and it earns its length by being the answer to all of them at once, every time, without you retyping.

Who you are and what you do. Role, organisation, the kind of work that fills your week. Two or three lines. This determines what the machine assumes you already know, which changes every explanation it gives you.

What you are working on now. The two or three live things. This is the part that goes stale, so keep it short and date it.

Who your output is for. Your team, your director, your clients, the public. Audience does more to shape a piece of writing than any style adjective, and it is the thing you are most likely to leave out because it is obvious to you.

How you want things written. Not adjectives. Length defaults, format defaults, whether you want the answer first or the reasoning first, words you never use. "British spelling, answer first, no bullet lists unless the items are genuinely parallel" is worth more than "professional but friendly". Module 07 turns this into a real standards file built from samples of your own writing; for now, a few concrete defaults.

What you never want back. The output shapes you always delete: the opening flattery, the summary of what you asked, the three-paragraph windup before the answer, the closing offer to help further.

Keep the whole thing under a page. This file gets read alongside every request you make, and a long one crowds out the material you actually want it to work from.

Key insight: about-me.md is not an introduction. It is the answer to every question you are tired of answering.

Lists of banned inputs age badly, because what is safe depends on the tool, the plan, the settings and the year. Learn a question instead, and ask it every time.

What happens to this if it ends up somewhere I did not choose?

Not "will it leak" — you cannot know that. The question is what the consequence would be, and the answer sorts most things quickly. A draft of your own team update, ending up somewhere unintended, is embarrassing at worst. A client's unreleased financials, someone's medical situation, a colleague's performance problem, a password, a contract under an NDA that names the parties: those have consequences that land on someone who did not choose this. That difference is the whole judgement.

Three follow-ups when the answer is not obvious. First, is this mine to share, or does it belong to a client, a patient, a candidate, a colleague? Second, could I get the same result with the identifying parts removed — names swapped, figures rounded, the company called "the client"? Usually yes, and redaction costs a minute. Third, does the setting I am using retain this and use it for training, and do I know that or am I assuming it? Check once for your tool, write the answer in a line in your file, and stop guessing.

If a piece of work fails all three, do it yourself. Module 00 already told you that some tasks are yours; this is another way of arriving at the same conclusion.

Persistent instructions apply to everything, and that is exactly the problem some of the time.

Three cases where you should turn them off. When you are drafting for someone else — a note that goes out in your director's name, a message in a colleague's voice — your standing style rules will fight the job and you will spend the session undoing your own defaults. When you want a genuinely outside reading of something, your context tells it what you think, and it produces the most plausible continuation of a conversation with a person who thinks that; Module 04 makes use of exactly this. And when you are working on something that has nothing to do with your job, the standing context is dead weight taking up room.

Know where the switch is. Being able to turn your setup off deliberately is part of having a setup.

• ai-kit/ exists on your own disk and contains both ai-map.md and about-me.md.
• Your file names a specific audience for your output, not "stakeholders".
• Your style section contains format and length defaults, not adjectives.
• A fresh conversation, with no explanation from you, produces something in your format for your audience.
• You can point at the surface holding your standing context and name it.

Open a blank file and answer the five questions in two lines each: what you do, what you are working on, who reads your output, what format you want answers in, and what you always delete, plus one line on what you have decided not to paste. Fifteen minutes with a timer beats a perfect file you never finish. The version you write by hand is usually more honest about your style than the version an interview produces, because you are not performing for the interviewer.

Your kit exists, and the file that describes you is in it and loaded. Every module from here on assumes this file is present and current, so put a note in your calendar to reread it in three months, when the "working on now" section will be wrong.

Save as: ai-kit/about-me.md — read again in Module 02, and in every module after it.

Say these out loud. If one stalls, reread the section named beside it.

• I can name my tool's five surfaces and which one holds my standing context. (The Five Surfaces)
• I can point at ai-kit/about-me.md on my own disk, show that it is under a page, and show where it is loaded. (What Goes in about-me.md)
• I can say what I would ask myself before pasting something sensitive. (What Not to Paste, and How to Decide)

Module 02, Outcome Over Instructions, takes the requests you are already making and rebuilds them as five-part briefs, using ai-kit/about-me.md as the part you no longer have to write out.