So far you've used AI one question at a time. Now learn to hand it something bigger — a whole project with many steps — and work through it together without losing the thread.
The same briefing instinct from Lesson 1 — pointed at something much bigger.
Ask AI to "write my business plan" or "build me a website" and you'll get something that looks substantial but misses half of what you meant. Not because AI is weak — because the request gave it nowhere to land.
It's the same lesson as your first conversation, just scaled up. A single-shot question works for a single-shot task. A project has many steps, decisions, and unknowns — and those have to be worked through, not guessed in one go.
Before you ask AI to do anything, tell it what "done" looks like. Four things, in plain sentences:
The goal. What should exist at the end that doesn't exist now?
The audience. Who is it for, and what do they expect?
The constraints. Length, tone, deadline, format, things to avoid.
What you already have. Notes, drafts, files, or decisions already made.
This is the single biggest upgrade. Instead of diving straight in, ask for the plan first:
Now you're editing a plan — cheap, fast, and easy — instead of fixing a finished draft you didn't want. You catch the misunderstanding before it multiplies across the whole project.
With a good plan approved, you now move through it in rounds. Each round is one step: you ask, review, correct, and move on. Three rules keep it clean:
One step per round
Don't move on until the current step is actually right. Errors compound if you stack steps on a weak foundation.
Correct in plain words
"Good, but shorter," "less formal," "keep this, redo that." You steer it like a colleague, not a vending machine.
Save good bits as you go
Copy approved pieces out into your own doc. Don't let a long chat be the only home for your work.
Keep the goal in view
If you drift, restate the goal and the plan. The plan is the map; re-read it when you lose your place.
The biggest way long projects go wrong is losing context: what you've decided, what you rejected, what the goal is. A few habits fix it:
Recap before continuing. When you come back later, paste a short summary of where you left off before asking for the next step.
Restate the goal periodically. It's the anchor that keeps a long project from slowly drifting off course.
Own the master copy. Your document is the source of truth — not the chat window.
Once you're comfortable working with AI in planned, multi-step rounds, you've basically learned how AI agents work. Agents are AI given tools and a goal — so they can carry out steps for you rather than just drafting each one.
The difference is who runs the loop. Today, you drive: you plan, you ask, you check. An agent can run more of that loop itself. But the skill you just practised — scoping, planning, stepping, verifying — is exactly what makes agents useful instead of dangerous to your work.
On a real team, the lead doesn't do every task — they set the goal, agree the plan, assign the steps, and sign off on the work. That's your role here too.
The more you act like a lead — clear goal, agreed plan, checked steps — the more AI behaves like a capable teammate instead of a confident stranger. Scale is just briefing done deliberately.
Your feedback shapes what we build next.
You can start a conversation, write a clear brief, pick the right tool, build the habit, check the work, and run a multi-step project with AI at your side.
Three lessons left — and they cover the things that separate people who use AI from people who get good work out of it.
Hand it an example instead of a description — and correct it like a colleague.
You've run the loop by hand — now see how agents run more of it for you.
Go behind the curtain and understand how all of this actually works.
Essential when a project's output is going to other people.
Back to the full tutorial library.
