Half the students we meet use ChatGPT for everything; the other half refuse to touch it. Both groups are wrong, in different ways.
Every time we visit a campus or meet students at a hackathon, we get a version of the same question. "How do you use ChatGPT for assignments?" Or sometimes: "Is using ChatGPT cheating?"
We've gotten it enough times that we figured we'd write it down.
From what we see, students fall into roughly two camps. About half use ChatGPT (or Claude, or whatever's free that week) for everything. The other half refuse to use it at all because their professor said it was "cheating." Both groups are missing the point, in different and instructive ways.
We can tell. We can always tell. Not because the writing is bad (it's often quite good) but because it has no fingerprints on it. There's no point at which the explanation gets clumsier because the author was thinking. There's no detail that's wrong in a revealing way. It reads like a textbook, and textbooks are not what makes someone hireable.
Here's the part we want you to hear: your job in school is not to produce the best possible artifact. Your job is to develop a brain that can produce artifacts after you graduate, when you don't have an instructor telling you what to write. If you outsource the thinking, the muscles don't grow. The artifact is fine. You won't be.
We respect the discipline. We don't respect the strategy.
You will graduate into a workplace where every single one of your colleagues uses these tools. You will be slower. Your work will look hand-crafted in a way that, in industry, reads as "stuck." Refusing to learn how the tool works is not the same as being principled. It's just being unprepared.
Use it after you've done the thinking, not instead of. The order is non-negotiable.
Write your terrible first draft. Solve the problem badly. Get something concrete on the page that came from your own head. Then bring in the model: to argue with you, to find what you missed, to point out where your explanation will lose a reader. The model is a brilliant editor and a mediocre author. Treat it that way.
You'll know you're doing it right when the model disagrees with you and you're sometimes correct. That's the moment your brain has caught up enough to push back. That's the moment we look for in interviews.
If you're using it for citations: stop. It hallucinates them. We've watched this end badly enough times that we don't have patience left for it. Use the library. The library still works.
Three years ago we had AGI timelines on a whiteboard. Today we keep a Notion doc tracking which of our predictions were wrong, and why the boring takes keep beating the spectacular ones.
Read Enterprise AIMost enterprise AI postmortems blame poor planning. Ours blames good planning. We rolled out without a strategy and accidentally found the right one.
Read Small BusinessMost SMB AI advice is written for companies with budgets they don't have. Here's the playbook we keep seeing small businesses use to outpace bigger competitors on stacks under $200/month.