Skip to main content
Plan mode helps you explore different ways to implement a feature. The AI researches how other products solve similar problems, studies your codebase to understand your constraints, asks clarifying questions, then proposes several implementation approaches — each with reasoning about why it works. When you approve a plan, it becomes a working branch you can see, test, and refine. Compare multiple approaches side by side and merge the best parts.

​
How Plan Mode Works

1

Research

Before asking questions, the AI automatically researches:
2

Study Your Project

The AI analyzes your codebase to understand your patterns, components, and architecture.
3

Ask Clarifying Questions

The AI asks questions to better understand your requirements and preferences.
4

Create Multiple Plans

The AI generates several implementation plans, each with a different approach or variation.
5

Review and Approve

Review the plans and select which ones you want to implement.
6

Create Branches

For each approved plan, the AI creates a new branch (frame) in your canvas, each implementing a different variation of the feature.

​
When to Use Plan Mode

Explore Approaches

When you want to see multiple ways to implement a feature

Product Decisions

When you need to compare different UX or architectural approaches

Learn from Others

When you want the AI to research how other apps solve similar problems

A/B Testing

When you want to create multiple variations to test

​
Setting Up Favorite Apps

To help the AI understand your design preferences:
  1. Browse screens at 21st.dev/community/screens
  2. Bookmark apps that inspire you or match your style
  3. The AI will reference these when researching solutions in Plan mode
The more apps you bookmark, the better the AI understands your taste and can suggest solutions that match your preferences.

​
Example Workflow

Imagine you need to add a user profile page. Here’s how Plan mode helps:
  1. You request: “Add a user profile page”
  2. AI researches: It looks at components from 21st.dev, studies your bookmarked apps (maybe Linear’s clean profiles, Stripe’s detailed views, Vercel’s minimal approach), and analyzes how similar products handle user profiles.
  3. AI asks: “Should it include an activity feed? What information is most important to your users?”
  4. AI creates 3 plans:
    • Plan A: Minimal profile with key stats (inspired by Vercel’s approach)
    • Plan B: Full profile with activity timeline (similar to Linear)
    • Plan C: Card-based layout with tabs (more traditional, easier to scan)
  5. You approve Plans A and C — you want to compare minimal vs. detailed
  6. AI creates 2 branches — each with working code implementing a different approach
  7. You compare visually in the canvas, see which feels right, and merge the best parts
You get working code for each approach, so you can make an informed decision by comparing real implementations.
After Plan mode creates branches, you can use Iterate mode to refine each implementation or merge ideas from different branches.