A year ago, Boris Cherny at Anthropic released Claude Code internally — a terminal-based AI agent that doesn’t just suggest code, but reads files, runs commands, and makes changes across your project. Boris claims the terminal UI (Text UI - TUI) agent pattern is here to stay.
GitHub Copilot CLI brings the same idea to our toolchain. Install it:
brew install copilot-cli
Then type copilot in any project directory to start a session.
It can read your code, edit files, run shell commands, and interact with GitHub — create PRs, work on issues, the lot. It reads your copilot-instructions.md and AGENTS.md (Tip #1) and supports plan mode (Tip #3). When it wants to run something, it asks permission first.
The shift from “AI suggests a line” to “AI works on a task” is the big one. Terminal agents are where that shift is happening fastest.
💡 Try this: Install Copilot CLI and start a session in a project. Ask it to explain the codebase before asking it to change anything.
Have you tried a terminal-based AI agent?
🟢 Yes — I use one regularly
🟡 Tried it, still figuring it out
🔴 Not yet, but curious
⚪ I prefer staying in the IDE
What’s been your experience?