Autopilot vs --allow-all in GitHub Copilot CLI

🤨 “What’s the difference between autopilot mode and –allow-all in GitHub Copilot CLI ?”

This is a common question I see all the time, and the documentation does a great job of explaining it.

But, here is a quick mental model:

🔓 –allow-all = “don’t ask me for permission” 🤐 –no-ask-user = “don’t ask me questions” 🤖 autopilot = “just go finish the job”

Expanding this a bit more:

🔓 –allow-all skips permission prompts. Copilot still stops at every decision point and waits for you.

🤐 –no-ask-user stops clarifying questions. But Copilot still won’t take multiple steps on its own.

🤖 Autopilot mode does both AND keeps going. It chains steps together autonomously until the task is done. No waiting for your input between rounds.

✨ Where autopilot really shines: Start in plan mode (Shift+Tab) → build a detailed plan with Copilot → then flip to autopilot to execute it.

Clear plan = better results. ✅

⚠️ Where it doesn’t: vague prompts, open-ended exploration, or anything needing your judgment along the way.

(super simple example below using Azure Cosmos DB vNext emulator with MCP)