Back to the RevenueCat homepage

Chapter 3: Build your first version

Find a problem worth solving

You don’t need a completely original idea. Many useful apps take something people already do and make it simpler, faster, or better suited to a specific audience.

One way to find opportunities is to study existing apps.

Search the App Store for the type of app you’re considering, then open three to five popular examples. Read their recent reviews and look for:

  • Problems that appear repeatedly
  • Features users keep requesting
  • Tasks that feel confusing or unnecessarily complicated
  • Complaints about pricing or what sits behind the paywall
  • Positive reviews that explain what users value most

Written reviews usually tell you more about the user experience than star ratings alone, making them a useful place to look for recurring problems and unmet needs.

Use AI to analyze reviews

You can copy a sample of relevant reviews into Replit Agent or another AI assistant and ask it to identify patterns.

For example:

Group these App Store reviews into recurring problems. For each problem, tell me:

1. How frequently it appears
2. Who appears to experience it
3. What outcome the user wants
4. A simple app feature that could address it

Base the answer only on the reviews provided. Don’t invent additional evidence.

Treat the result as a starting point, not proof. Read the reviews yourself and make sure the patterns the AI identifies are actually present in the source material.

Define the smallest useful version

Before asking the Agent to build anything, decide on the one thing somebody should be able to accomplish with your app.

Think through:

  • The user: Who is the app for?
  • The problem: What are they struggling with?
  • The main action: What should they be able to do?
  • The result: What useful outcome should they receive?
  • What can wait: Which features don’t belong in the first version?

Start with one focused, phone-friendly flow rather than trying to build the complete product immediately.

Replit Agent, or another AI assistant, can help you work through these questions before you start building.

Build in Replit

From the Replit home screen:

  1. Describe the app you want to create
  2. Select Mobile app as the app type, or specify this in your prompt
  3. Start in Plan mode if you want to brainstorm with Replit Agent and map out the work before it makes changes
  4. Review the proposed plan and correct anything that doesn’t match your idea
  5. Start building

Plan mode lets you discuss the approach and review the proposed tasks before Replit Agent starts changing the project. Use it to check that the Agent has understood what you want before approving the plan.

Write a useful first prompt

You don’t need to explain how to write the code. Your job is to describe the product clearly and explain what success looks like.

A useful first prompt should cover:

  • Who the app is for
  • The problem it solves
  • The main user flow
  • What the first version must include
  • What it should not include yet
  • How the app should look and feel

If you’re not sure how to structure the prompt, you can ask an AI assistant to help you write it.

Improve the app one change at a time

The first version probably won’t be exactly what you imagined. That’s normal.

Vibe coding works best as a short loop: ask the Agent to build something, try it yourself, identify the most important problem, ask for one focused change, then test it again.

Avoid vague prompts such as:

“Make the app better.”

Instead, explain what you want to change and what Replit should preserve.

For example:

Simplify the planner screen so users can scan the whole week more easily. Keep the existing navigation, but reduce the amount of text on the recipe screen. Don’t change anything in the Shopping menu.

When something breaks, describe what you did, what you expected to happen, and what actually happened.

For example:

When I open an individual recipe and select ‘Add to planner’, the loading indicator appears but never disappears. I’d expect it to disappear once the recipe has been added to the planner. Fix that flow.

For larger changes, return to Plan mode and agree on the approach before the Agent edits the app.

Use checkpoints when a change goes wrong

Replit Agent creates checkpoints as it works. Think of them as save points you can return to if a change makes the app worse.

Use focused feedback when the result is close. Roll back to a checkpoint when:

  • A previously working feature breaks
  • The Agent changes several unrelated parts of the app
  • The new version is clearly worse
  • Fixing the change would be harder than returning to the last working version

Don’t spend an hour trying to prompt your way out of a bad change if you can return to the last version that worked.