Vibe coding in prod

Erik Schulntz discusses "Vibe coding" and how to responsibly implement it in production, defining it as coding where one fully trusts AI-generated code, emphasizing the exponential growth in AI task capabilities and the need to adapt. He draws an analogy to compilers, advocating for trusting the system while focusing on product outcomes rather than the underlying code. Schulntz addresses concerns about managing implementations without full understanding, suggesting focusing Vibe coding on leaf nodes in the codebase to contain potential tech debt. He advises acting as a product manager for the AI, providing comprehensive guidance, and highlights a successful 22,000-line change to their production reinforcement learning code base written heavily by Claude. The talk concludes with a Q&A session, addressing topics such as learning in the age of AI, balancing information input, cybersecurity, test-driven development, and practical coding workflows.