GitHub Copilot CLI adds second-opinion review

This is a smart move: agentic coding tools are most dangerous when they’re confidently wrong, and a second model family is a practical way to catch blind spots without turning the workflow into committee software.

• –The review happens at high-leverage checkpoints, especially after planning, after complex implementations, and after test writing.
• –GitHub’s own evaluation claims Rubber Duck closes a large share of the gap between Claude Sonnet and Claude Opus on hard multi-file tasks, which is a meaningful signal even if the benchmark is narrow.
• –The feature is still experimental, so the real question is latency and interruption cost versus the quality of the catches it adds.
• –This also shows where Copilot CLI is headed: not just code generation, but agent supervision and quality control inside the terminal.
• –For teams already using CLI-based agents, cross-model critique could become the default guardrail for refactors and high-risk changes.