Comparing Claude 3.5 vs 3.7: Building an Expense Sharing App

A Tale of Two AIs: Developing the Same App with Different Approaches

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.