AI Agent Product Experiment
An experimental project designed to evaluate the real-world capability of AI coding agents in building a complex, animation-heavy, multi-feature product with minimal manual intervention.
Overview
Designed a feature-rich Valentine’s interactive app and delegated implementation primarily to an AI coding agent.
Evaluated AI performance across UI generation, animation logic, state management, and responsiveness.
Minimized direct code writing to supervision, debugging, architectural correction, and prompt iteration.
Role
Human Supervisor / Product Architect
Duration
1 hour
Platform
Web
Stack
React (Vite), Tailwind CSS, Framer Motion, Supabase
The Challenge
Can modern coding agents build cohesive, production-level front-end applications?
How much human intervention is required for animation-heavy, stateful apps?
Where does AI break: architecture, performance, edge cases, or design coherence?
- AI-generated code often lacked structural consistency across modules.
- Animations required refinement beyond initial generation.
- State persistence logic needed manual architectural correction.
- Performance optimization was not handled autonomously.
Process & Timeline
1 hour
Designing, Building, Launch
Mostly build using Claude Code
Key Screens
Home page
Our story page
Love letter page
Reasons page
Bucket list page
Virtual gift page
Results & Impact
~95%
AI Code Contribution
Initial draft generation
~10% of codebase
Manual Refactor
Architecture + bug fixes
8
Feature Modules
All functional with supabase backend
Key Learnings
AI excels at generating small, modular components but is weak at designing long-term system architecture.
Prompt quality directly influences structural integrity.
AI-generated animation code often requires manual performance tuning.
Supervision and reading code is more valuable than direct coding in AI-augmented workflows.
- The future developer is an AI orchestrator.
- Specification clarity > coding speed.
- Architecture remains a human advantage.
- AI increases leverage, not autonomy.
Interested in working together?
I'm open to new projects and collaborations.