An AI just wrote code with me for a week. It's a brilliant intern, not a replacement.
GitHub Copilot suggests whole functions as you type — and "will AI replace developers?" is trending again. After a week with it, the honest answer is more useful than the panic.
⌨️ An AI just wrote code with me for a week. It's a brilliant intern, not a replacement.
GitHub's Copilot preview has been the thing developers can't stop arguing about — an AI, built on OpenAI's Codex, that suggests whole functions as you type. And right on cue, the old headline is trending again: will this replace programmers?
I spent a week with it. My honest read: wrong question.
What it actually is, is the most capable junior teammate I've worked alongside — fast, eager, widely-read, and confidently wrong just often enough that you can never stop reviewing its work. It is astonishing at boilerplate, at the function you've written fifty times, at reminding you how an unfamiliar API is called. It is unreliable exactly where judgment matters — the edge case, the security implication, the "why" behind the code.
Which is why "replace" is the wrong frame. The value isn't automation; it's acceleration of the tedious, so humans spend more of the day on the parts that need a human. A brilliant intern doesn't replace the senior engineer — it changes what the senior engineer spends their time on.
Two things stood out that matter beyond coding:
This is GPT-3's family, pointed at a real job. Those language models we've watched as research curiosities are quietly becoming tools that do work. That transition — from demo to daily driver — is the thing to watch, and it will not stay confined to code.
The bottleneck moves to review. When generating a first draft becomes free, the scarce skill becomes judging it — knowing what's right, what's subtly wrong, what to throw away. That's a more senior skill than typing, not a less one.
The instinct to panic or to dismiss are both lazy. The useful posture with a tool like this is the one you'd take with a talented new hire: give it the grunt work, check everything, and let it free you up for the judgment only you can bring.
The question I'd put to any team eyeing tools like this: not "will it replace us," but "which half of our work is the boilerplate we'd happily hand to an intern — and are we spending enough of our time on the other half?"