A product manager built Opener Studio to automate cold outreach, starting with 'vibe coding' using Lovable. Initial speed turned to chaos as AI-generated code created fragile architectures, race conditions, and localStorage misuse. Debugging became a time sink, revealing AI's tendency for subtle, hard-to-trace errors that compound complexity.

He switched to Cursor for better control, then adopted a Roguelite development methodology inspired by video games. Instead of debugging, he treats failed runs as learning opportunities: run fast, declare 'death' when complexity peaks, extract lessons, and rebuild from scratch using git restore. This turns code into disposable prototypes while preserving institutional knowledge.

The final structure is a Synthetic Team: himself as CEO, Gemini as CTO for architecture reviews, and Cursor as lead engineer. He enforces rules via .cursorrules, mandating a 'thick backend, thin client' design. This disciplined, iterative approach aims to harness AI's speed without sacrificing maintainability or creating technical debt.