All posts

Comparing Claude 3.5 and 3.7: building an expense sharing app

Two versions of Claude got the same brief for a Flutter expense sharing app. They took remarkably different paths.

  • ai
  • claude
  • flutter
Two robots representing Claude 3.5 and 3.7 playing Rock 'em Sock 'em Robots, generated by ChatGPT

This unique blog post documents how two different versions of Claude AI approached building the same Flutter-based expense sharing application. The fascinating part? They took remarkably different development paths despite receiving the same instructions.


🤖 Claude 3.5’s Approach

As observed in conversation with Claude 3.5, here’s how it tackled the project:

Project Naming and Initial Setup

  • Conducted a thorough name brainstorming session
  • Checked Play Store availability for multiple options
  • Settled on “DivvyUp” with proper Flutter naming convention (divvy_up)
  • Focused on initial requirements gathering

Technical Architecture

  • Started with basic auth-related files
  • More gradual, step-by-step approach
  • Files created: auth_service.dart, auth_provider.dart, login_screen.dart

Implementation Style

  • More conversational approach
  • Focus on understanding requirements first
  • Gradual implementation strategy
  • Less initial code, more planning

🤖 Claude 3.7’s Approach

Based on the conversation with Claude 3.7, here’s how it approached the same project:

Project Setup and Structure

  • Started with “MoochNoMore” name without extensive discussion
  • Immediately established a complete directory structure:
    • models
    • screens
    • services
    • utils
    • widgets
    • providers

Technical Implementation

  • Comprehensive directory structure from the start
  • Immediate Firebase integration
  • Full implementation of user pairing system
  • Complex state management with Provider

Development Philosophy

  • Action-oriented approach
  • Rapid feature implementation
  • Strong focus on code quality (proactive lint fixing)
  • Extensive codebase development from the start

Feature Focus

  • Advanced user pairing system
  • Complex state management with Provider
  • Comprehensive Firebase integration
  • Enhanced UI with settings screen and navigation

Code Quality Practices

  • Proactive lint error fixing
  • Proper naming conventions
  • Clean code practices
  • Detailed documentation of changes

👤 Cascade’s Analysis

As a third AI system examining both approaches, I notice some fascinating differences in development philosophy:

Development Spectrum

These two approaches represent opposite ends of a development spectrum:

  • Planning-Heavy (Claude 3.5): Prioritizes understanding requirements and creating a solid foundation before diving into code
  • Implementation-Heavy (Claude 3.7): Focuses on rapidly building comprehensive technical solutions

Strengths & Weaknesses Analysis

Claude 3.5’s Approach:

  • ✅ Better user requirement understanding
  • ✅ More methodical, reducing potential rework
  • ✅ Stronger product naming process
  • ❌ Slower initial progress
  • ❌ Less comprehensive technical solution

Claude 3.7’s Approach:

  • ✅ Faster feature delivery
  • ✅ More complete technical implementation
  • ✅ Better code quality practices
  • ❌ Potentially rushed requirements phase
  • ❌ May build features that aren’t needed

Real-World Implications

In professional development environments:

  • Claude 3.5’s approach aligns with agile methodologies that emphasize iterative development and continuous user feedback
  • Claude 3.7’s approach resembles feature-driven development that focuses on building complete feature sets

The Ideal Hybrid

The most effective approach would combine:

  1. Claude 3.5’s thoughtful requirements gathering and user-centered focus
  2. Claude 3.7’s technical excellence and quality practices
  3. A balanced implementation pace that allows for both planning and execution

Note: This multi-perspective analysis is based on partial conversations and represents an educational comparison rather than a definitive assessment of either AI system’s capabilities.