GitHub expanded Rubber Duck in Copilot CLI on May 7, 2026. Rubber Duck is Copilot CLI’s cross-family review agent, intended to give a second opinion from a different model family.
When a Copilot CLI session uses a GPT model as the orchestrator and /experimental is enabled, Copilot can dispatch a Claude-powered Rubber Duck critic. When Claude is the orchestrator, GitHub says the critic model can now be GPT-5.5.
Why this matters
The interesting part is not the brand mix. It is the admission that single-model coding agents miss things a different model family may catch: architecture mistakes, subtle bugs, and cross-file conflicts.
This is the direction serious AI coding tools are moving: orchestration plus independent review, not one model generating code and grading itself.
Buyer take
Developers using Copilot CLI should test Rubber Duck on real refactors, not only toy tasks. The best fit is likely code where a second opinion is worth the latency: migration scripts, security-sensitive changes, database logic, and multi-file edits.
For teams comparing Copilot with Cursor, Claude Code, and Codex, Rubber Duck improves Copilot’s CLI story but does not replace human review. Treat it as a risk-reduction layer before PR review.
What is still unclear
The feature is behind /experimental, and GitHub did not publish quantitative evals showing how often Rubber Duck catches issues missed by the orchestrator.
Sources
Primary and corroborating references used for this news item.