Objective
Your mission
Use the plan-first prompt pattern to get an architecture plan from the AI before it builds. Review, approve/modify/reject, then build. Output: plan_first_build.md.
You need: Any AI coding tool. A text editor.
Time: 30 minutes.
Step 1: write a spec with the plan-first sentence (5 min)
Write a spec for a small app using the 5-part framework from Lesson 01. Then add this sentence at the end:
"Before you write any code, outline your approach: what files you'll create, what each file does, what technologies you'll use, and what the page structure will be. Wait for my approval before building."
This transforms the interaction from spec → code to spec → plan → (your review) → code.
Write in your file:
## My Spec
[Your 5-part spec]
## Plan-First Sentence
[The sentence above, pasted at the end of your spec]
Step 2: review the AI's plan (10 min)
The AI should return something like:
Approach:
index.html — single page with all contentstyles.css — all styling, mobile-firstapp.js — app logic
Structure:
- Header with title
- Input form
- List display
- Data in localStorage
Technologies: Vanilla HTML/CSS/JS. No frameworks. No backend.
Check for:
- Scope — Building too much? Too little?
- Technology choices — Using a framework when you wanted vanilla? A database when you wanted localStorage?
- Page structure — Layout matches what you described?
- File organization — Makes sense for the project size?
Write in your file:
## AI's Plan
[paste or summarize the plan]
## My Review
- Scope: [right / too much / too little]
- Technology: [right / wrong — what would you change?]
- Structure: [right / needs changes]
- File organization: [right / needs changes]
Step 3: approve, modify, or reject (5 min)
- Approve: "Looks good, go ahead and build."
- Modify: "Close, but two changes: (1) use localStorage, not a database. (2) Put everything in a single index.html file. Then build."
- Reject: "This isn't what I described. I said single-page with localStorage, but you're planning a multi-page app with a database. Reread my spec and try again."
Write in your file:
## My Decision: [Approve / Modify / Reject]
[What you said to the AI]
The cost of skipping the plan
When you skip the plan step, the AI makes architecture decisions without your input. If those decisions are wrong, fixing them means asking the AI to restructure — harder than building from scratch. A 2-minute plan review catches architecture issues before they become code. The plan step is the cheapest correction point in the entire build cycle.
Step 4: build and compare (10 min)
Let the AI build after your approval/modification. Compare the result to what you expected.
Write in your file:
## Result
Did the plan review catch anything that would have been harder to fix after building? [yes/no — what?]
Did the build match the plan? [yes/no]
What would have gone wrong if you'd skipped the plan step? [specific scenario]
The AI's plan for your localStorage-only list app includes a database and three pages. What's the cheapest fix?
The plan is the cheapest correction point. Fixing architecture after the code exists means restructuring, which is harder than building from scratch.