A live demo walkthrough of the 'rubber duck' feature in GitHub Copilot CLI, which automatically requests code reviews from a competing AI model family (e.g., GPT reviewing Claude's plan) at key boundaries like planning, implementation, and test generation. The presenter explains that combining Claude and GPT model families yields better results than using either alone — approximating ~75% of Opus performance using Sonnet plus GPT review at lower cost. The session also covers the REPL loop (called 'Ralph loop') concept, where an agent iteratively implements a PRD and stores learnings across iterations, and gives beginner tips for getting started with the Copilot CLI using plan mode and autopilot.
•16m watch time
150 Impressions