if you’re building with AI agents, you need to read this article.   most CLIs were designed for humans.   agents can’t press arrow keys, answer interactive prompts, or guess what your flags do this is a clean breakdown of the patterns that actually work:  → make every input passed as a flag. no interactive prompts as the default path → let agents discover commands step by step instead of dumping all docs upfront → every –help needs real examples. agents pattern match off examples faster than descriptions → accept stdin and flags for everything. agents think in pipelines → fail fast with errors that actually tell you what to fix → make commands idempotent. running the same command twice shouldn’t break anything → add –dry-run for anything destructive → return useful output on success, not just emojis this applies way beyond CLIs. any interface you’re building that an agent will touch needs to be designed for agents, not just humans the era of building apps, products, and tools for humans only is ending build for both or you’ll get automated around