By ByteBot
GitHub Copilot CLI shipped four new features at Build 2026 on June 2 — and unlike most “AI updates,” these are worth actually reading about. Rubber Duck (a cross-model code critic), local voice input, scheduled background prompts, and a redesigned terminal UI are all live now on every paid Copilot tier. The rubber duck is the one to pay attention to.
Rubber Duck: A Critic From a Different AI Family
Rubber Duck is a built-in review agent that runs alongside your main Copilot CLI session. After drafting a plan, mid-implementation on something complex, or after writing tests, the primary agent hands its work to the rubber duck for a second opinion. The rubber duck looks for blind spots, design flaws, and overlooked edge cases. It reports back with concrete, actionable feedback — it doesn’t make file changes. Your main agent decides what to do with the feedback.
The architectural choice that makes this interesting is the cross-model design. If you’re running Claude as your session model, the rubber duck uses GPT-5.4 as the critic. If you’re using GPT, it uses Claude. GitHub’s reasoning is direct:
“A model reviewing its own output can only catch what its training allows it to see. A model from a different family brings different assumptions, different blind spots, and different strengths.”
That’s the right call. Single-model self-review is fundamentally limited — the same training biases that caused the mistake will prevent the model from catching it. A critic from a different model family doesn’t share those failure modes.
Performance data backs this up. According to GitHub, Claude Sonnet with Rubber Duck closes 74.7% of the performance gap between Sonnet and Opus on real-world tasks. On the hardest problems, it delivers a +4.8% improvement. A real example from the docs: Rubber Duck caught three files all reading from a Redis key the new code stopped writing. The confirmation UI and cleanup paths would have silently broken on deploy — exactly the kind of subtle cross-file dependency a single model reviewing its own work is likely to miss.
The name comes from rubber duck debugging — the classic technique from The Pragmatic Programmer (1999): explain your code to an inanimate rubber duck, and the act of explaining reveals the bug. GitHub’s version upgrades the duck from passive to active. Instead of talking at an object, you get a critic that talks back — from a different AI lineage. You can invoke it manually with /rubber-duck.
Voice Input: On-Device, Private, Generally Available
Voice input is now generally available in GitHub Copilot CLI, and the implementation detail that matters most is where the transcription runs: locally. Audio never leaves your machine. GitHub uses the built-in speech recognition APIs on Windows and macOS — no API call to a cloud transcription service.
To use it: hold Space and talk. Or press Ctrl+X → V to start recording, speak, then press any key to stop. The transcribed text inserts as your prompt. English and Spanish are supported. The voice runtime downloads on first use; your preferences persist in ~/.copilot/settings.json across sessions.
Local-only transcription should be the default for developer tools. Cloud transcription adds latency, privacy risks, and API dependencies. On-device transcription is faster, private by design, and works offline. Copilot CLI ships this at GA where other major coding CLIs don’t offer it at all.
Scheduled Prompts: An Ambient Agent in Your Terminal
Two new slash commands — /every and /after — let you schedule prompts within a session. /every runs a prompt on a recurring interval; /after fires once after a delay.
# Poll for new review comments every 30 minutes
/every 30 check for new PR review comments and resolve them
# Run tests once, 2 hours from now
/after 2h run the test suite and summarize any failures
# Open the schedule manager
/everySchedules are session-scoped — they only run while your Copilot CLI session is open. The minimum interval is 10 seconds, the maximum is 24 hours. For persistent scheduling, wrap a Copilot CLI invocation in cron or Windows Task Scheduler. Full scheduling docs are here.
The practical applications are real: autonomous PR review monitoring, CI failure investigation, periodic linting checks. This turns Copilot CLI from a reactive Q&A tool into something closer to an ambient agent watching your repo and acting on it.
Redesigned Terminal UI
The new terminal interface (experimental) adds tabs for Session, Issues, Pull Requests, and Gists — press Tab to switch in a repo context. Theme-aware semantic colors include default, github, dim, high-contrast, and colorblind modes. The layout adapts to narrow terminals without truncating content, and screen reader support auto-enables when detected.
What’s Actually Available Right Now
- Rubber Duck: Generally available on all paid Copilot tiers (Free, Pro, Pro+, Max, Business, Enterprise)
- Voice Input: Generally available, requires GitHub CLI v2.20.0+
- Prompt Scheduling: Experimental — enable with
/experimental on - New Terminal UI: Experimental — enable with
/experimental on
Copilot CLI is becoming a development control surface — not just a terminal chatbot. The same week GitHub also shipped the Copilot App (a dedicated desktop agent manager, currently in technical preview), which signals where this is heading: Copilot as a persistent background environment you work alongside rather than occasionally query. Full release notes on the GitHub Changelog.
