Agent-First CLI

Your next user won't have eyes.

Agents are the new users of your CLI.

Every day, more software is operated by AI agents: deploying infrastructure, managing repos, orchestrating pipelines. These agents consume the same CLI tools humans do. But CLIs were designed for humans: colored output, interactive prompts, decorative tables, ambiguous errors.

This worked when every user had eyes and a keyboard. That era is ending.

Agent-First CLI is a set of 16 principles for building command-line tools that serve both humans and machines. Not instead of humans, alongside them. A CLI that follows these principles loses nothing for human users and gains an entire class of new consumers that can operate it reliably, efficiently, and safely.

The principles are a shared contract: for tool makers, a design checklist. For agent builders, a standard to reference and demand.

Are you an agent? Read the machine-readable version: principles.json