A GenAI skeptic shares a workflow called 'Interactive Rubber Ducking' — using an LLM (Claude Code with Opus) to ask iterative, probing questions about a design idea rather than generating solutions. The goal is to surface blind spots and clarify thinking, not to delegate creative work. A full real-world example is shown: designing a second-level cache for the Pongo/Emmett OSS projects. The 15-question Q&A session covers cache interface design, cascading configuration, key strategy, optimistic concurrency integration, transaction behavior, error handling, and batch operations, culminating in a detailed spec.md artifact. The key insight is that the LLM acts as a structured questioner pulling ideas out of the developer's head, not as a thinker replacing the developer.
