GitHub Copilot Rubber Duck is a new review feature in GitHub Copilot CLI that uses a second, different model family to challenge the plan, implementation, or tests produced by the primary coding agent. The core insight is that AI coding agents often fail not with obvious errors but with polished, internally consistent output built on flawed premises. Self-review by the same model is limited, so introducing a different model perspective at key checkpoints can catch expensive mistakes before they propagate across multi-file changes. The author argues this signals a healthier direction for AI-assisted engineering — one built around generation, review, and deliberate challenge rather than ever-increasing autonomy — but cautions that the feature only stays valuable if it remains targeted and avoids becoming noisy ceremony.

6m read timeFrom thomasthornton.cloud
Post cover image
Table of contents
Why a second opinion actually helpsWhat starts to matter on harder tasksWhy this matters more at platform scaleIt only works if it stays targetedA better direction for AI-assisted engineering

Sort: