Why Unix is Perfect for LLMs

The tools that feel oldest are the ones AI works best with.

Every coding agent session generates GitHub activity as a byproduct (commits, PRs, issues).

I’m seeing teams use GitHub Issues more than traditional PM tools.

This makes me think: modern software could benefit from Unix style. Composable. Text-based. CLI-first.

Infrastructure software already works this way.

Terraform, Docker, Kubernetes—all text-based, composable, CLI-first. That’s why AI agents work seamlessly with them.

Why Unix is perfect for LLMs:

• LLMs are text machines: Unix commands speak the same language. No pixels, no mouse, no visual state, no GUI (Text I/O is genuinely easier for LLMs than pixels)
• Stateless execution: Doesn’t need to remember what menu is open and what is selected. Input → Output
• There is a lot of Unix training data

As Cursor and Claude Code adoption grows, I expect every modern software company will eventually add CLI support.

Not for developers. For AI agents.