Below are six chains of instructions. Each one carries a single real task to the end: not a lesson example — your files, your spreadsheet, your archive.
Copy one step at a time, paste it into the agent, read what comes back. You do not need the course for this: if it works, you will see for yourself why to go further. If it does not, you spent twenty minutes and learned where the edge is.
Why hands first, rather than "let me read up on it". Reading about AI gives orientation, not skill: you can scroll the channels for a year and end up where you started. Confidence grows from what you did, not from what you read — which is why there is not a single line of theory here.
Why this is not programming
The tool has the word "code" in its name, and that word has cost people a lot of hours: they decided it was not for them. It is not an editor plugin. It is an agent that sits on top of your folders — it reads files, renames them, rewrites them, runs things and reports back.
None of what follows is development. It is admin work you are still doing by hand.
🗂️The folder where nothing can be found
Work routine · ~20 min
Four hundred files named "scan_final_2 (copy).pdf". Finding one means opening half of them.
you need: a folder of files (a copy, to be safe)
Step 1
Look through the files in this folder and show me a table: current name, what it actually contains, what a clear name would be. Do not rename anything yet.
First you see what the agent understood. This is where you catch "it decided these were contracts when they are invoices".
Step 2
Naming scheme: YYYY-MM-DD_counterparty_document-type. Rewrite your proposed names to match it and show the table again.
You set the scheme. Without one the agent invents its own — pretty, and inconvenient for exactly the way you search.
Step 3
Good. Rename according to that table, but first save the current names into rename-log.csv so this can be undone.
The rename log is your undo button. You have to ask for it before, not after.
Step 4
List the files where you could not determine a date or counterparty, and leave those untouched.
The leftovers are the valuable part: the ten files you will still open yourself — but ten, not four hundred.
What you end up with: A folder where the filename answers "what is this", an undo log, and a short list of the doubtful ones.
For twenty files, doing it by hand is faster. This chain pays off from a hundred up.
✉️Ninety letters that each need to be personal
Work routine · ~30 min
You have a spreadsheet of people. Each needs a human letter, not a blast — but you do not have ninety letters in you.
you need: a spreadsheet (CSV or export) and one letter written by you
Step 1
Open this file and tell me what columns it has and which of them could make a letter different for each person. Do not write anything yet.
Sometimes there is nothing to personalise with: just a name and an email. Better to learn that up front.
Step 2
Here is a letter I wrote to one person: [paste your letter]. Break it down: what here is common to everyone, and what belongs to that person specifically.
You hand over your voice as the sample instead of asking "write a letter". The difference is audible in the result.
Step 3
Draft three letters — for the first three rows. Keep the common part word for word, build the personal part from their data. Show me; do not send.
Three is the size where you can still spot the fake note. At ninety you will not proofread it.
Step 4
Fixes: [what is off]. Apply them and produce the rest, each as its own file in letters/. I will send them myself.
Sending stays with you. An agent holding your mail credentials is a separate decision, not a first-day one.
What you end up with: Ninety drafts in your own voice, left for you to read and send.
If the personalisation is just inserting a name, this is a mail merge and the agent adds nothing.
🧾Two exports that should agree
Work routine · ~25 min
A payment-processor export and a bank statement. Once a month you reconcile them by eye and never catch everything.
you need: two files of transactions covering the same period
Step 1
Here are two files. Do not reconcile yet. Tell me which fields could match them at all, and how their date and amount formats differ.
Ninety per cent of discrepancies are date formats and fees, not missing money.
Step 2
Match transactions by amount and date with a three-day tolerance. Show me three lists: matched, left-only, right-only.
The tolerance matters: money settles with a delay, and exact date matching produces a pile of false mismatches.
Step 3
For each unmatched row, propose an explanation: fee, refund, delay, duplicate. Where you are unsure, say so.
"Say where you are unsure" turns a confident tone into an honest list. Without that line you get tidy explanations for everything.
What you end up with: A short list of rows that genuinely need your decision, instead of an hour spent between two spreadsheets.
This is triage, not accounting. The result is still checked by whoever is responsible for the books.
🧠The pile of notes you avoid opening
Personal · ~25 min
Years of notes: fragments, links, thoughts on the run. You remember writing it down somewhere and cannot find it.
you need: a folder of notes (export from any app)
Step 1
Read these notes and name the themes that recur. Do not rename or move anything yet. Do not invent tidy categories — name them the way I name them myself.
Someone else’s taxonomy is the main reason a second brain dies. Asking for your own words changes the outcome.
Step 2
Take one theme — [name] — and assemble the scattered notes into one coherent text: what I already understood, where I contradict myself, what is missing.
"Where I contradict myself" is the reason to do this at all. You can see the list of notes yourself; the contradictions you cannot.
Step 3
Now sort the notes into folders by theme — copies, leave the originals alone. And list the ones that fit nowhere.
Copies, not moves: change your note system only once you are convinced the new shape is better.
What you end up with: One living text on a theme that matters to you, and a sense of what is actually in the archive.
The agent will produce a coherent text even where there is no coherence. Read it as a draft from a partner, not as a conclusion.
🔍A subject you must understand by Friday
Personal · ~20 min
An unfamiliar field: a medical report, a contract, an industry new to you. Googling means drowning; there is no one to ask.
you need: a document, or just the name of the subject
Step 1
Explain this subject to someone who knows nothing about it but is not stupid. Three levels: in two sentences, in a page, and what people usually get wrong here.
"What people usually get wrong" is the most useful of the three, and almost nobody asks for it.
Step 2
Now give me five questions to ask a specialist so that I cannot be misled.
The goal is not to replace the specialist but to stop being helpless in the conversation.
Step 3
What is all that based on? Mark separately: what is settled, what is disputed, and what you filled in yourself.
This question separates knowledge from plausible fluent speech. Always ask it — especially when you like the answer.
What you end up with: Enough grasp to hold a conversation with a specialist, plus a list of what to verify.
On medicine, law and money the agent speaks just as confidently as on anything else. Confidence means nothing here — verify.
📁The household document archive
Personal · ~30 min
Bills, contracts, certificates, warranties — across three places and a camera roll. When you urgently need one, it cannot be found.
you need: a folder of scans and photos of documents
Step 1
Go through these files and build an inventory: what the document is, what it relates to, whether it expires. Do not touch the files.
The inventory is useful on its own even if you stop here: for the first time you see the whole archive as a list.
Step 2
Flag in the inventory what expires soon or has already expired, and where I appear to have no second copy.
This is the step that justifies the rest: an expired policy is found by inventory, not by search.
Step 3
Sort copies into folders: home, health, transport, work, other. Leave the originals where they are; put the doubtful ones aside.
Personal documents are not where you trust automation on the first run. Copies and a separate "doubtful" pile handle that.
What you end up with: An inventory of the household archive, a list of what needs renewing, and sorted copies.
Here you are showing the agent passports and contracts. Before starting, decide where it runs and where the data goes. In the course that is the first conversation, and not by accident.
Where this will waste your time
The honest part, without which the rest would be advertising.
The first run takes longer than it looks. Installing, granting folder access, working out where it reads files from — that is an evening, not five minutes. The second time takes minutes.
Small volumes do not pay off. Twenty files, three letters, one page — faster by hand. It starts to matter when the count runs into hundreds.
Taste-driven work comes out average. Where your judgement is needed — what to cut, whom to refuse, what to call it in your own words — you get an averaged answer and spend the time redoing it.
It is wrong with full confidence. The tone is identical whether it is right or filling gaps. That is why every chain starts with "show me, change nothing": skip that and you will be undoing things.
Your data goes wherever the model runs. For a household archive or work documents that is its own conversation: which tool, running where, what leaves your machine.
If it worked
Then it was never about ability or about the word "code". The course is the same thing carried further: your own stack, your own chains for your own work, and an agent that holds your context instead of starting from zero every time.
If it did not work, that is an answer too, and an honest one. You spent an evening and found the tool’s edge — more than any course description could give you.