You've never written code. That's exactly why you should be the one running these agents.

Let me describe your last week with AI, and tell me if I get it close.

A report you needed to make sense of, an email you kept not sending, a pile of notes you were supposed to turn into something useful. So you opened the ChatGPT app, or Claude, whichever one you’ve settled into by now, and you typed your question into the little box, and you pasted in some context, and it thought for a second and handed you something back.

Then, you copied the good bits out of the chat, you pasted them into your real document, you fixed the formatting, you renamed the file, you went and found the thing it referenced to make sure it actually existed, and you stitched the whole thing together yourself, by hand.

There’s a newer way to work now, though, and the people who’ve found it don’t really go back. It isn’t a smarter chat. It’s a different surface entirely, one where the model doesn’t advise on the work, it does the work. You give it a goal, point it at a folder, and it reads the files, makes a plan, runs for as long as the job takes, and hands back finished files with a record of what it changed. Millions of people already work this way.

I want to scratch the surface of “coding agents” today. The popular ones are Codex from OpenAI, and Claude Code from Anthropic.

And if you’ve never written a line of code, here’s the part that matters: this is more yours than it is the engineers’. Knowledge work is files in, files out. Spreadsheets, decks, transcripts, the messy export, the folder nobody’s opened in a year. You drive these agents in plain English, the way you’d scope a job for someone competent.

That’s the case I’m making today.

Below, what it looks like when it runs, why the non-coders are the ones who win here, and the one five-minute thing you can try tonight. And very soon I’m following all of it with a full, step-by-step guide to getting the most out of coding agents, written specifically for people who are starting out.