I Stopped Chatting With Claude Code and Started Engineering With It

Your terminal isn’t a chat window. It’s an operating system for AI-assisted development, and most people haven’t figured that out yet.

Nuno Roberto | Chief AI and Data Architect

The first time I used Claude Code, I treated it like a very expensive chat window. I’d type a question, read the answer, type another question, read that answer, and at some point during the evening I’d realize I’d been having a conversation with my terminal for three hours and had produced roughly the same output I could’ve gotten from a well-structured Google search. That’s the trap most people fall into. They install Claude Code, run it, interact with it the way they interact with every other AI tool they’ve used, and then wonder why the experience doesn’t feel transformative.

It isn’t transformative yet because they’re using it wrong. Not morally wrong, not technically wrong in the sense that it throws errors, but architecturally wrong. They’re treating a cognitive system like a conversational one. And that distinction, once you see it, changes everything about how you work.

I figured this out the hard way, over roughly six months of building Kairos, an enterprise agentic AI platform, almost entirely through Claude Code. In that time, I went from ad-hoc prompting to a…