AI Crash Course, call 2 of 5. Monday, Nov 9.

Talk
to it

Last week you gave it a brain.
This week: how to ask.

Jaden Parisi

New chat

Sure! Here's a complete lead scraper for Toronto. I'll pull business listings from a directory site and save everything to a CSV.

Let me know if you'd like pagination, proxies, or a Google Sheets export!

The bad ask

Ten words. No brief. Typed cold.

Looks smart.
Useless.

The AI wasn't dumb.

The ask was empty.

What it is

A power tool. Not a mind reader.

A drill is a great tool. It still doesn't know where you want the hole.

0xAI=0

10xAI=0

It multiplies what you bring.
Bring nothing, get nothing, fast.

What it can see

What you typed Build me a lead scraper. I want leads from Toronto.

Your city

Your niche

What the leads are for

Your budget

What you tried last week

What "good" looks like

It only sees the tip.

Why it does thatEvery blank you leave, it fills with the most typical answer on the internet. Vague in, average out.

The cheat code

You don't need to be good at prompting.
You need to answer questions.

Here's what I want: [your idea].
Before you do anything, interview me. Ask every question you
need answered to do this well. Tell me what decisions I'm not seeing.
Don't take my words literally. Figure out what I'm actually trying
to achieve, and tell me if what I asked for won't get me there.

It knows the questions. You know the answers.

Writing it yourself

Six questions. Answer them and the brief writes itself.

What do you want, and why?The reason changes the output.
Who is it for?"Contractors, 50s, skeptical, on a phone."
Where should it look?Your handbook, a file, a site, a screenshot.
What does "done" look like?Format, length, what it has to match.
What can't change?The rules, and what to leave alone.
What's your hunch?And tell it to prove you wrong.

What do I want, and why?
Who is it for?
Where should it look? (my handbook, files, sites, screenshots)
What does "done" look like?
What can't change?
What's my hunch? (and tell it to prove me wrong)

0:14. Explainer, fast. For when you write the brief yourself instead of being interviewed. Don't read all six; point at three.

Number 1, the why. Story: "never use long sentences" fixed the one spot it thought I meant. "Keep sentences short, this gets read on a phone by a busy contractor" fixed the whole document, headlines and buttons included. A bare rule is a pattern to match. A reason is a goal to serve. It can only generalize from the second one.

Number 4, "done": the one everyone skips, and the one next week's build is built on. Number 6, the hunch: the bottleneck audit story. "I think it's a closing problem. If it's a different problem, what is it?" It came back with a different answer. That's how you stop it agreeing with you.

Hands: the copy button drops the six as a checklist. Suggest they save it as a sticky note in their brain folder.

Always another way

When it says it can't, it's out of ideas.
Not out of options.

That didn't work. Give me 5 completely different ways to get the same result. Include the weird ones.

Tunnel vision

Same fix, third time. It's stuck.

Other tells: "let me try one more thing," every attempt gets more complicated, it keeps saying sorry.

Why it does thatEverything it already wrote steers what it writes next. It can't un-see its own wrong turn. A fresh chat can.

Step back

Stop. Forget your approach. What are 3 totally different ways a pro would do this?

Start fresh

Two misses means a new chat and a better brief.
Don't keep patching.

0:20. Explainer. The train rolls on its own. One track, a wall coming, and it will not change tracks by itself. If you see the same fix twice, it's stuck.

Story: the ad batch took six rounds. The first five were the same approach with patches on top. Every round got more complicated and it kept saying sorry.

The why: everything it already wrote is now part of its instructions. It's defending its own first guess without knowing it. That's why "try again" in the same chat gets worse, and why a fresh chat with a better brief gets better. Press: the two moves. Step back: make it throw its approach away and list three totally different ones. Start fresh: after two misses, stop correcting. New chat, better brief, include what you learned. Your brain folder makes the fresh chat cheap: it already knows your business.

Feedback that works

My launch video, two tries.

Try one

"Kinda boring, feels like a PowerPoint."

Nothing changed.

Try two

"One continuous shot, Apple style, song starts at 0:40, use this sound folder, landscape."

"Really good."

Feelings don't help it. Say what, where, and why.

Stop doing these

No magic words. Just a clear ask.

Today's examples. The list changes every year. The rule doesn't.

ALL CAPS and "CRITICAL"
Threats and fake tips
"Act as a world-class expert"
"Think step by step"
Magic templates from the internet

0:26. Explainer. Fast. The strikes run on their own. Lead with the rule, it's on screen: no magic words, just a clear ask. Then one line per example as they get crossed out.

Caps and "CRITICAL": current models over-react to it. Threats and tips: no measurable effect. "Act as a world-class expert": doesn't make it smarter; a one-line role is fine for tone, that's all. "Think step by step": the models already do this on their own now. Templates from the internet: a clear brief beats a clever template.

"That's the first half. Everything so far works in any chat window. Now we go to your Claude Code, because it can read your brain." Swap at 28:00. Jaden becomes Hands (drives Claude Code, narrates every decision). Co-presenter becomes Explainer (runs the slides, keeps time, reads the chat).

Second half, follow along

Now use it on your real work.

1
Pick one task
From your scoreboard. Something you'd do this week anyway.
2
Run the interview
Answer its questions out loud, in the chat. Short answers are fine.
3
Turn it into a brief
Read the brief. Fix one line if it's wrong. Then say go.

Read my handbook and scoreboard.md first. The task I want done: [the task].
Before you do anything, interview me. Ask every question you need answered to do this well. Tell me what decisions I'm not seeing. Don't take my words literally. Figure out what I'm actually after.

Turn my answers into a one-page brief: what I want and why, who it's for, where to look, what "done" looks like, what can't change, my hunch. Save it as brief.md and wait for my OK before you start.

0:28 to 0:48. Hands: Jaden. Explainer: co-presenter. Share the screen. Claude Code open in the brain folder. "Everyone do this with me, on your own task. Hands up in the chat when your interview starts."

Hands: pick one real task off your scoreboard, out loud. Paste the first prompt with the task filled in. Read its questions out loud and answer them fast, the way you'd answer a new hire. Explainer points at what it asks that Jaden never said: "That question is the iceberg. That was under the water."

Paste the second prompt. Read the brief on screen, fast. Fix one line ("change X to Y"). Narrate why. Then "OK, go." While it works, Explainer reads the chat and takes two questions. Watch for people whose Claude didn't read the handbook: tell them to name the file.

Timing: 6 minutes to pick and interview, 4 to brief, then it runs. If a student's interview is better than yours, put theirs on screen.

Shortcut for anyone set up through Claude From Zero: the kit ships /interview-me, which is this prompt saved as a skill. Mention it once; keep the pasted prompt on screen because it works in any AI.

Mini exercise, 8 minutes

Catch it being wrong.

It did the task. Before you use a word of it:

Before I use this: what did you guess? What did you assume I meant? Where did each fact come from? Which part would a pro push back on?

When you find one: point at the line, say what you expected, run it again.

Your ship this week

One real task, start to finish.
Interview it. Brief it. Post the before and after.

The best one goes on screen next Monday.

Monday, Nov 16

Build with it.

Idea to a working tool.

Jaden Parisi

Talkto it

A power tool. Not a mind reader.

You don't need to be good at prompting.You need to answer questions.

Six questions. Answer them and the brief writes itself.

When it says it can't, it's out of ideas.Not out of options.

Same fix, third time. It's stuck.

Step back

Start fresh

No magic words. Just a clear ask.

Now use it on your real work.

Pick one task

Run the interview

Turn it into a brief

Catch it being wrong.

One real task, start to finish.Interview it. Brief it. Post the before and after.

Notes

Talk
to it

You don't need to be good at prompting.
You need to answer questions.

When it says it can't, it's out of ideas.
Not out of options.

One real task, start to finish.
Interview it. Brief it. Post the before and after.