The Incremental Build
Don't ask for everything at once. Build up step by step.
The Problem with Big Asks
When you ask for a complete feature in one prompt:
- AI makes assumptions you didn't want
- Harder to review large chunks of code
- Mistakes compound through the whole thing
- Difficult to backtrack if something's wrong
The Solution: Step by Step
Start with the foundation
"Create the database schema for a todo app with users and todos"Review. Approve. Move on.
Add the next layer
"Now add the API routes for CRUD operations on todos"Review. Approve. Move on.
Continue building
"Add the React components to display a list of todos"Then interactivity
"Add the form to create new todos"Finally, polish
"Add optimistic updates when creating todos"Benefits
| Benefit | Why It Matters |
|---|---|
| Reviewable chunks | You can actually read and understand each piece |
| Early error catching | Fix issues before they propagate |
| Better AI context | AI has clearer picture of what exists |
| Easier debugging | Know exactly which step broke things |
| Cleaner commits | Each step = one commit |
When to Use
Use incremental builds for anything that takes more than ~5 minutes or touches multiple files.
✅ Good for Incremental
- New features
- Database schema + API + UI
- Complex refactors
- Multi-file changes
❌ Just Ask Directly
- Single file changes
- Bug fixes
- Simple additions
- One-liner tweaks
Example: Building a Chat Feature
Step 1: "Create the database schema for messages (sender, receiver,
content, timestamp, read status)"
Step 2: "Add API routes: GET /messages/:conversationId, POST /messages"
Step 3: "Create the MessageList component to display messages"
Step 4: "Add the MessageInput component with send button"
Step 5: "Wire up real-time updates using WebSockets"
Step 6: "Add read receipts"Each step builds on the last. Each step can be reviewed and tested.