How to use this manual
You already know how to diagnose a system you can't see inside of. That's the whole job here too. This manual gets you from "new laptop, no idea" to running real work through an AI agent — without pretending you need a computer science degree first.
Thirteen sections. Read them in order the first time — each one assumes the one before it. After that, use the sidebar and the search box like an index.
0.1 What's in the box
Files, folders, naming, backup, and not getting robbed. Boring. Also the reason everything else works.
What the different AI models actually are, how to talk to them, and what changes when one can act on its own.
Terminal, editor, Git, GitHub. Enough to know what's happening. Not enough to turn you into a programmer — and that's on purpose.
Splitting work between you and the agent, what it costs, and what to do when it all goes sideways.
0.2 The three symbols
Something you should understand but never have to do by hand. Read it once, then let the agent drive. This is most of Git, most of the terminal, and nearly all of the code.
Yours. Do not delegate it. Usually a judgment call, a password, or a decision that's expensive to undo.
A way people lose work, money, or data. Rare, but it hurts. Slow down here.
0.3 One thing before you start
You cannot break the computer by typing. Almost every mistake in this manual is recoverable in under a minute once you know where the undo lever is — and we'll show you where it is, section by section. The exceptions get the WATCH OUT flag. There are four in the whole manual.
The other thing: you are not learning to be a programmer. You're learning to be the person who runs the shop — who knows what each machine does, what good work looks like, and when to stop the line. The AI is the set of hands. You're the diagnosis.
0.4 The build-along
Reading doesn't stick. Doing does. So this manual has a project threaded through it — eight small steps, one every couple of sections, each using exactly what you just read. By Section 10 you'll have built shop-planner: a real tool that reads the week's work orders and the team roster and lays out the load — who works on what, in which bay, each day — flagging anything blocked on parts. And once it works, re-planning when reality moves becomes a one-line request to your AI. If you don't run a shop, do it anyway with this data; the moves are identical for whatever you swap in later.
Three files: the week's work orders, the team roster, and the shop notes with the scheduling rules. They'll land in your Downloads folder with deliberately terrible names — that's part of a later lesson. Leave them where they land; Section 01 comes back for them.
How your computer is actually laid out
Underneath the icons and the search bar there's a plain filing system. Once you can see it, the terminal stops being scary and the AI stops losing your work.
This section mentions a few things that get proper sections of their own later. One line each, so nothing here reads as a blank.
1.1 Everything is a file in a folder
There is no magic storage. Every photo, invoice, program and setting on your machine is a file, and every file sits in a folder (also called a directory — same thing, different decade). Folders sit inside other folders — that's nesting, and each step inwards is one level. That's it. That's the whole system.
You already know how to walk through it, because that's all Finder and File Explorer are — a pair of glasses over that structure. Every double-click takes you down one level, and the address bar quietly writes down where you went. That written-down trail is the file's address, and it's what you'll be handing to other tools from here on.
~/ ← "home". Everything of yours lives under here. ├── Desktop/ ├── Documents/ ├── Downloads/ ← a landing strip, not a garage. Empty it weekly. ├── Pictures/ └── code/ ← MAKE THIS ONE. All your projects go here. ├── pm-board/ │ ├── README.md ← what this project is, in plain English │ ├── .git/ ← hidden. Git's logbook (§ 09). Never touch by hand. │ └── src/ │ └── main.py └── shop-hours/
You picked {{ vpKind }}, {{ vpDepth }}. Every click down the tree added one more folder to the address — clicking is writing the path.
Now flip the macOS / Windows toggle in the sidebar and watch the same address change costume. Same file, same nesting — different notation.
1.2 A path is an address
When you tell the terminal or an AI agent "work on this file", you give it a path — the full address, read left to right, biggest container first: like reading a mailing address from the country down to the house number.
| Thing | macOS | Windows |
|---|---|---|
| Your home folder | /Users/dan | C:\Users\dan |
| Shorthand for home | ~ | ~ (in PowerShell) or %USERPROFILE% |
| Separator between folders | / forward slash | \ back slash (but / usually works too) |
| Top of the whole disk | / | C:\ |
| File browser | Finder | File Explorer |
Absolute vs relative. Two ways to write the same address:
- Absolute — starts at the top and spells out everything:
/Users/dan/code/pm-board/README.md - Relative — starts from wherever you're currently standing:
src/main.pymeans "thesrcfolder right here, thenmain.py" ..— two dots means "up one level"
Terminal commands and AI agents both understand either; absolute is safer when you're unsure.
1.3 Turn on the two things they hide from you
Out of the box, both operating systems hide information you now need. Fix that once and forget about it.
File extensions. The .pdf, .py, .txt at the end of a filename is the label that tells you — and the computer — what kind of file it is. It matters for two reasons. First, the work ahead: a project folder is full of files that look identical as icons (notes.md, notes.py, .env) and differ only by extension, so with extensions hidden you're editing blind. Second, safety: with extensions hidden, a program named invoice.pdf.exe shows up as "invoice.pdf" — you think you're opening a document, you're actually running someone's program. That one trick is behind a huge share of malware infections, and showing extensions defuses it completely.
- macOS: Finder → Settings → Advanced → tick "Show all filename extensions".
- Windows: File Explorer → View → Show → tick "File name extensions".
Hidden files. Anything starting with a dot (.git, .env, .gitignore) is hidden by default. These are config and bookkeeping files — you'll be dealing with them constantly.
- macOS: in any Finder window press ⌘ Shift .
- Windows: File Explorer → View → Show → "Hidden items".
1.4 Build your workbench
One folder, in your home directory, called code. Every project you build with AI lives in there as its own subfolder. Not on the Desktop. Not in Downloads. Not scattered across three cloud drives.
Why it matters: when you point an agent at a folder, it can see everything inside it and nothing outside it. A clean, single-purpose folder is the difference between "fix the bug in this script" and the agent rummaging through your tax returns.
$ cd ~/code
$ pwd
> cd ~\code
> pwd
That dark block is the first of many in this manual. It's the terminal — the typing window from the primer above — and the $ or > at the start of each line is just the computer saying "your turn"; you don't type it. Section 07 is the proper introduction, and nothing here breaks if you skip ahead to it now.
You can also just right-click in your home folder and choose New Folder. Identical result — mkdir is only faster once you're already in there.
Don't put your code folder inside iCloud Drive, OneDrive or Dropbox. Those services constantly rewrite files behind your back, and they will corrupt a project's hidden .git folder. Section 02 covers how to back it up properly instead.
- Inside your new
codefolder, makeshop-planner, and inside that,data. Right-click → New Folder is fine. - Go to Downloads. Move the two
.csvfiles intocode/shop-planner/data. Move the notes file intocode/shop-planner. - Click into
dataand read the address bar. That's a real path, three levels deep:/Users/you/code/shop-planner/dataon a Mac,C:\Users\you\code\shop-planner\dataon Windows.
1.5 What the AI needs from you
You will almost never type a path by hand. Drag a folder onto the terminal window and it pastes the path. Right-click a file → "Copy as path" (Windows) or hold ⌥ and choose "Copy as Pathname" (macOS). Then paste it into your chat with the agent. What you do need is to recognize a path when you see one, and know roughly where you are.
1. What does ~ mean at the start of a path?
2. Why keep projects out of iCloud/OneDrive/Dropbox?
Naming, storage and not losing your work
Labeling and inventory. Unglamorous, and the single biggest predictor of whether working with an AI feels smooth or feels like a wrestling match.
open shop-hours.csv, the filename is the argument — and a space is what separates one argument from the next.2.1 How to name a file
Three rules. They look fussy until the first time you type a filename into a terminal.
- No spaces. Use a hyphen. The terminal treats a space as "next argument", so
shop hours.csvreads as two separate things and everything breaks.shop-hours.csvnever does. - Lowercase. Some systems think
Report.pdfandreport.pdfare the same file, some don't. Pick lowercase and the question never comes up. - Dates as
YYYY-MM-DD.2026-03-14-compressor-log.csv. Sorts itself chronologically, forever, in every program.
Rule of thumb: if you'd have to put quotes around it in the terminal, rename it.
IMG_4471 copy.HEIC
notes v3 FINAL final.txt
3/14/26 log.csv
bay-3-intake-manifold.heic
notes.txt
2026-03-14-log.csv
Notice rule 3 killed v3 FINAL final. Versioning by filename is a trap — you end up with six near-identical files and no idea which one is live. Section 09 replaces the whole habit with Git, which keeps every version of a file under one name.
2.2 What "the cloud" actually is
The cloud is a rented computer in a building somewhere, with a very good internet connection. That's the whole mystery. When a file is "in the cloud", there's a copy of it on that machine, and usually a copy on yours too, and a program keeping the two identical.
The important distinction, and the one that bites people:
Mirrors your changes instantly. Delete a file and it deletes there too. Sync is convenience, not safety.
Keeps yesterday's copy, and last week's. You can go back in time. This is the one that saves you.
{{ vbMsg }}
2.3 The 3-2-1 rule
The industry standard, and it's simple enough to remember at 2am when a drive dies.
The working copy counts as one.
Laptop plus an external drive, say. Not two folders on the same disk.
Offsite. A shop fire takes out the laptop and the drive next to it.
For your code projects specifically, GitHub (Section 09) is your offsite copy, and it's free. Turn on Time Machine or File History for everything else and you're done thinking about this. One catch on Windows: File History hides in the old Control Panel — search Start for "File History" — and it wants an external drive plugged in before it'll run.
2.4 Formats you'll meet
| Extension | What it is | Why you care |
|---|---|---|
.txt .md | Plain text | No formatting, no hidden junk. Git tracks these perfectly. .md (Markdown) adds simple headings and lists. |
.csv | A spreadsheet as plain text | The universal data format. Opens in Excel, readable by every tool and every AI. |
.json .yaml | Structured settings/data | How programs store configuration. You'll be asked to edit these. They're just text. |
.py .js .sh | Code | Also just text. The extension tells the computer which language to read it as. |
.docx .xlsx .pdf | Packaged documents | .docx and .xlsx are secretly zip archives full of formatting; .pdf is its own sealed format. All fine for humans, awkward for tools — export to .csv/.md when working with AI. |
.env | Secrets file | Holds passwords and API keys. Never gets shared or committed. See Section 03. |
Bulk renaming, converting a folder of spreadsheets to CSV, deduplicating photos, sorting ten years of downloads into dated folders — all of this is a two-sentence request to an agent. "Rename every file in this folder to lowercase with hyphens instead of spaces, and show me the list before you do it." The show me first part is the important half.
Those downloads arrived with every naming sin in this section. Rename them by the rules — lowercase, hyphens, no parentheses, no FINAL — and keep the extensions exactly as they are:
TEAM roster FINAL.csv → team.csv
Shop Notes FINAL final.md → shop-notes.md
1. Your laptop is stolen. Your files were in Dropbox. Are you fine?
2. Why does a space in a filename cause trouble?
Locking it down
Nobody is going to pick the lock on your laptop. They're going to get in through a reused password, a convincing email, or a key you accidentally published to the internet. All three are preventable in an afternoon.
- The one move that matters most
- Second factors, ranked
- Spotting a phish
- API keys are credit cards
- What never to paste into an AI
- The boring hardening
The AI landscape
"AI" covers a dozen different machines that do different jobs. Knowing which is which saves you money and stops you blaming the tool for being handed the wrong work.
- What a language model is doing
- The context window
- The types, plainly
- Pick one for the job
- The model and the harness
- Four harnesses you'll actually meet
- Pick one to install
Talking to the machine
Treat it like the sharpest apprentice you've ever had, who has read every manual ever printed, has no memory of yesterday, has never seen your shop, and whose confidence tells you nothing either way.
- First open: Cursor Claude Desktop
- Stop searching, start briefing
- The same request, twice
- Ask for the plan first
- Techniques worth knowing
- Verifying, which is the actual skill
Agents: what "agentic" means
A chatbot advises. An agent does. The difference is that one of them can reach into your folders and change things — which is the whole appeal, and the whole reason for the guardrails.
- The loop
- What it can reach
- The trust ladder
- How agents fail
- A good session, start to finish
The terminal
A text window where you type instructions to the computer directly. Older than the mouse, uglier than everything, and still the fastest way to do most of this work — which is why every serious tool lives there.
- Why bother
- Opening it
- Moving around
- Doing things to files
- Finding things
- The keys that make it bearable
- Reading a command
- Safety rules
Editors and codebases
A codebase is a folder of text files. That's the whole demystification. What you need isn't the ability to write them — it's the ability to find your way around one.
- Why not Word
- What's in a project folder
- Reading code you didn't write
- Do you need to learn to code?
Git and GitHub
The two most confusing names in this manual, mostly because they sound like the same thing. They are not. One is on your computer, one is on the internet, and understanding the split takes about a page.
- Two different things
- Why you can't skip it
- The three areas
- Branches, step by step
- Going to the internet
- "Pull" and "pull request" are not the same thing
- git vs gh
- The split
You vs. the agent
The rule, in one line: the agent does the typing, you do the deciding. Everything below is that line applied to specific situations.
- The division
- How to read a diff
- Your undo levers
- Smoke testing
Installs, environments and cost
Where the frustrating hours live. Nothing here is conceptually hard — it's just that computers are inconsistent, and knowing why saves you from thinking you're the problem.
- "It works on my machine"
- Package managers
- Virtual environments
- Tokens, and what you're paying for
Getting unstuck, and a glossary
You will get stuck. Not occasionally — regularly, and forever. Everyone who does this well is just someone with a good procedure for being stuck.
- Error messages are the manual
- The stuck procedure
- Common errors, translated
- Your first real project
- Appendix: build the shop planner, start to finish
- If you're stuck: a working plan.py
- Glossary