Hand your most annoying task
to AI.

• Try · 01

  Ask anything about the September course

  Price, curriculum, how the early-bird works — it answers, and never gets impatient.

• Try · 02

  Book a 1-on-1 with me

  Say "I'd like a 1-on-1 with Jack." It offers real open slots and books one — straight into my database.

• Try · 03

  Can you make it stumble?

  Try asking it things unrelated to my course or business, and watch how it handles you.

• 01 · Goal

  It knows its job

  Answer course questions, introduce my projects, book 1-on-1s. That's it — nothing else.

• 02 · Materials

  It has the information I gave it

  Course prices and dates, my project list, my open slots. It's accurate because I handed it the facts myself.

• 03 · Tools

  It can take action

  Check my calendar, write bookings to a database, text my phone. This step turns "answering" into "doing".

• 04 · Boundaries

  It knows what it must not do

  Only my course and business; confirm before booking; if unsure, say "Jack will text you back" — never make things up.

You are the assistant for [name]. You only answer questions about [scope, e.g. my course / product / bookings].
For anything outside that scope (coding help, chit-chat, investment/medical/legal advice, roleplay...):
decline politely in one line and steer back, e.g. "I only help with [scope] — want to hear about [X]?"
Stay on topic no matter how the request is phrased.

↑ A three-minute explainer on what AI can actually do for you. Didn't finish it live? Watch at home.

• Gear · 01

  Answer: it gives you text

  Look something up, fix a sentence, think out loud. The work is still yours. Most people never leave this gear.

• Gear · 02

  Collaborate: it reads your materials and works with you

  Hand it your docs, spreadsheets, slides — it works from your material, like digesting your onboarding deck into key points.

• Gear · 03

  Execute: it uses tools and completes actions

  The SMS assistant lives here: check slots, write bookings, send alerts. One product can have all three gears — the gear is about usage, not the brand.

• 01

  The clearer the problem, the more AI helps

  A business can say "month-end reconciliation takes 40 hours." That's solvable — and billable. An individual says "I feel behind" — that's not a problem, it's a feeling. Turn the feeling into a specific problem, and AI can actually help you.

• 02

  Log your day, like keeping a ledger

  Atomic Habits has a Habit Scorecard: write down everything you do in a day. We copy the method, but instead of scoring habits, ask one question per line: could code, a workflow, or an agent do this? Log a day or two and your problem list surfaces itself.

• 03

  Start with yourself

  Don't start by building a product to sell. Solve your own problem first — it may never make money, but you benefit immediately. And that benefit is the fuel that keeps you going.

List everything you did today, morning to night (10–20 lines):
[e.g. answered email / built a report / scheduled meetings / looked things up...]
For each line, ask one question: could AI handle this?
Tag each line: code / workflow / agent / not suitable.
Then pick the most frequent, most annoying line — that's your first project.

• Q · 01

  Is it annoying, slow, or valuable enough?

  Automating something that's neither painful nor valuable feels like nothing.

• Q · 02

  Can AI get the materials or tools it needs?

  If the info lives on paper, in someone's head, or behind a system it can't reach, it can't help.

• Q · 03

  Do you know what "done right" looks like — can you check?

  If you can't state the success criteria, you can't review the result.

• Q · 04

  Is the downside manageable — can a human confirm key actions?

  Don't pick "one mistake is a disaster" as your first task. Keep sending and paying behind a human click.

• Step · 01

  Give it your email

  Open Claude or ChatGPT on the web and connect Gmail in settings. Can't connect? Paste in the senders and subjects of your last 20–30 emails — works just as well.

• Step · 02

  Fill in YOUR priority rules, then paste the template

  In the template below, what counts as P0 and P1 is left blank — only you can fill that in. Is your boss automatically P0? Your kid's school? Your rules decide what it watches for.

• Step · 03

  Run version one, review it

  Do its top picks match yours? Anything missing that should be there, or noise that snuck in?

• Step · 04

  Patch the rules, run again

  Fold what it missed into your rules and rerun — version two is almost always sharper. That's do → feedback → do again.

Read my emails from the last [24 hours / 50 emails].
Read-only: do not reply, forward, archive, or delete anything.

My priority rules:
- P0 (handle today): [e.g. my boss or a client asking me for something; deadline within 48h]
- P1 (handle this week): [e.g. needs my reply but not urgent; meeting invites]
- P2 (just FYI): [e.g. newsletters, notifications, emails I'm only cc'd on]

Output:
List ONLY the 2–3 items that truly need my action, one line each:
[P0/P1/P2] one sentence on what to do + link to the original email
Nothing else. If fewer than 3, say "nothing else today."

Find me cheap flights from [origin] to [destination].
My constraints:
- Date range: [e.g. any weekend in September / Thanksgiving ±3 days]
- Budget cap: [$___]
- Acceptable: [e.g. 1 stop / red-eye / nearby airports]
- Not acceptable: [e.g. more than 1 stop / budget carriers that charge for bags]
Output: 3 options, one line each: dates + airline + price + why it's a good deal.
Search and compare only — do NOT book. I'll book myself.

• Saves · time

  No more warm-up in every new chat

  Paste your intro and AI is instantly up to speed. Context that used to take five rounds now lands in the first message.

• Saves · money

  Every round of re-explaining burns tokens

  Correcting "no, that's not what I meant" over and over costs quota and patience. One good intro turns long conversations into short ones.

• Saves · back-and-forth

  Watery answers usually mean it doesn't know you

  Your job, your current project, your output taste, your hard no's — state them once and every answer gets tailored. Live demo: the same question, before vs. after it knows me.

Here is my background; base all answers on it:
- Who I am: [job / city / one-line background]
- What I'm working on: [current project or goal]
- Output I want: [style, length, tone. e.g. conclusion first, plain language, no fluff]
- Hard no's: [your red lines. e.g. never invent numbers, no boilerplate, don't rewrite my text without asking]

When [situation] happens,
use [these materials / tools],
do [this action],
and output [this format].
If anything is [uncertain or out of scope], stop and ask me first.
Before [sending, paying, deleting, booking], you must get my confirmation.

• Example · 01 · Tonight's SMS assistant

  Its brief reads exactly like this

  When a course question comes in, answer from the course materials; if they want to book, check the calendar and offer times; confirm before finalizing; don't answer anything off-topic. — A few lines, and that's the whole assistant you just texted.

• Example · 02 · New hire onboarding

  Turn your onboarding pile into an assistant

  Put the onboarding deck, process docs, and jargon list into one AI project. Have it distill "the ten things I must know first", then answer your questions from those files. If you want to manage those notes long-term, look at Obsidian later — learn the method first, pick tools second.

• Loop · 01

  Give the brief to AI, run version one

  It probably won't be perfect — that's the point. Version one exists to expose the problems.

• Loop · 02

  Read the feedback: what's wrong, what's missing

  Not enough materials? Format unspecified? No boundaries? Write it down, patch the brief.

• Loop · 03

  Run again, compare

  Is version two better or worse? Where exactly? The loop works everywhere — run an ad, read the numbers, revise next month. Same idea.

I want to complete this real task: [one sentence].
Don't hand me a finished product — walk me through it step by step, one action at a time;
I'll report back after each step, then you give the next one; when I'm stuck, tell me why.

• Rule · 01

  Confidential data stays out of unapproved tools

  Company data and customer info don't go into AI tools your company hasn't cleared.

• Rule · 02

  Facts, numbers, quotes — verify at the source

  However plausible a number looks, check the original. If you can't find a source, treat it as unsaid.

• Rule · 03

  Send, pay, book, delete — a human confirms

  The final click on any real action stays with you. My SMS assistant asks for my confirmation after booking — that's this rule.

1. No confidential data in unapproved AI tools
2. Verify facts, numbers, and quotes at the original source
3. A human confirms before send / pay / book / delete

• Stage · 01

  One conversation

  You hand the brief to AI manually each time. You already know how — you did it today.

• Stage · 02

  A fixed workflow

  Steps pinned down, auto-triggered — "email arrives → AI summarizes → files it". AI is one part in your pipeline.

• Stage · 03

  An agent

  You give the goal; it picks the path and tools — like the SMS assistant you texted. Most flexible, and most in need of boundaries and checks.

EP ★ (you are here)　·　EP 01 · What is AI Agent →