OpenAI Codex

Talking to code

OpenAI Codex is an AI system that writes code when we give it prompts in plain English. We type what we want—“make a function that sorts numbers”—and she replies with code. No jargon, no compiler gymnastics. Just words in, code out.

The art of the prompt

The hard part is us, not her. If we’re vague, she’s vague. “Make a game” leaves her guessing. “Make a two-player tic-tac-toe game in Python” gets us something we can run. The rule of thumb: be specific enough that a human junior developer would know what to do.

Code as conversation

She isn’t a mind reader. We ask. She answers. Then we adjust. We can say “shorter function names” or “add comments” and she tries again. Think of it like pair programming with someone who never sleeps, but sometimes misunderstands.

When she shines

She’s best at everyday code—functions, small apps, glue between libraries. If we know the shape of the solution but not the exact syntax, she saves us time. She’s not great at entire systems. Asking her to “build Facebook” is like asking a cashier for financial advice.

Our takeaway

We prompt, she codes. We refine, she adjusts. That’s the loop. We still need to read what she gives us, because bugs slip through. But for learning, scaffolding, and getting unstuck, she’s useful.

And maybe that’s enough. We don’t need magic, just a partner who types faster than we do.