Chapter 01 / 06 · 3 min
Write a real briefing
A vague request gets a vague answer. I force myself to answer four questions before I hit send: who it should be, what the context is, what I want at the end, and in what form.
- Role: who it plays (a proofreader, a coach, an accountant).
- Context: what it needs to know, and what it must not touch.
- Goal: the result you want to see at the end.
- Format: a table, five lines, an email ready to send.
Make me a table of my customers.
You are my admin assistant. Here is my customer file (paste it). I want to see the customers who haven't paid, sorted by how late they are, in a table with three columns: name, amount, days late. Don't modify the original file.
Everyday uses to practice today: rewrite a tricky email, summarize a ten-page document, prepare questions before a meeting, compare two quotes, turn notes into a task list.
Chapter 02 / 06 · 3 min
Ask for proof, and the date
"It's done" proves nothing. An AI can tell you a job is finished when it isn't, with the same confidence as when it's right.
Does it work?
Don't tell me it works. Run it and paste me the result.
Same for things that move fast: menus, prices, quotas. An AI can describe a setting that no longer exists. Before a technical step, make it read today's official documentation.
Before any technical change, read the most recent official documentation online, and tell me the date of what you read.
Chapter 03 / 06 · 3 min
Set it once and for all
Fixing the same thing every week? Stop. There are two places for that, and they do two different jobs.
- Instructions: what you want every time (tone, language, format). In Claude: Settings > Your Claude, Instructions section. In ChatGPT: Settings > Personalization > Custom Instructions.
- Memory: what the AI keeps from your chats. In Claude: Settings > Your Claude, "Manage" next to Your memory. You read what it kept there, and you fix it.
Stop putting emojis everywhere.
Remember this for next time: no emojis in what I publish.
Chapter 04 / 06 · 2 min
Work with a backup
Before letting an AI change your files, keep a copy. If it breaks, you go back.
Something's broken, fix it.
Show me all the files you changed, then put them back the way they were.
Chapter 05 / 06 · 2 min
Keep your secrets
Password, card code, API key: none of that in the chat. The AI can tell you where to type it. You're the one who types it.
- Step 5.1: Pasted a key by mistake? Go to the service that gave it to you.
- Step 5.2: Revoke the old key.
- Step 5.3: Create a new one, and keep it out of the chat.
Chapter 06 / 06 · 3 min
Keep some friction
This one goes against the grain: don't hand everything over. If the AI does it all, you lose your grip bit by bit, and one day you can't judge what it gives back. So I keep some effort for myself on purpose.
My sorting rule: Back / Front. Back is what nobody sees: the AI sorts, searches, structures. Front is what people read with my name on it: I decide, I sign.
- Back: my morning brief. It gathers the info, files it, builds me the table. If it gets a line wrong, I see it and fix it.
- Front: my Insta captions. I write them myself. It's my voice, and if I hand it to the AI, it ends up sounding like everyone else's.