
AI Risks, Vibe Coding, and Refactoring

Technical Coaching in the Age of AI with Samman (Ensemble) - Emily Bache
Topic 7 of 8
Published Aug 25, 2025 · Clip 43:53–52:41 (8:48)
Emily discusses the risks of vibe coding: developers shipping AI-generated code they haven't reviewed undermines accountability for quality. She distinguishes prototypes/proofs of concept (fine for bulk AI code) from valuable, complex production codebases, where unintegrated, unmaintainable AI code is a recipe for disaster. She argues the fundamentals remain the same—modular design, small steps, describing problems concisely, verifying output, code reading, and refactoring skills (TDD in essence). She strongly criticizes AI for refactoring, citing studies showing it introduces bugs, dead code, and random changes, and advocates reliable deterministic refactoring tools instead, while acknowledging their usability problems (obscure menu names, cursor placement).