No AI slop.

Your output is your responsibility. A quick note on AI etiquette.

What's going on

Making something that looks polished is basically free now. Working out whether it's any good is not. That gap is the whole thing.

Polish used to be a decent signal that someone had put the time in. Not anymore. A five-second prompt and a week of careful work can come out looking the same.

The thinking and the checking still have to happen somewhere though. So when you forward something you haven't really read, you're just handing that bit to whoever's on the other end. You haven't done the work, you've moved it.

Sending something is a claim. So is posting it. So is submitting it. You're saying you've actually looked at the thing, and you're willing to put your name on it.

What slop looks like

• The strategy doc dump. Someone knocks out a 15-page plan with AI, barely reads it, forwards it with "thoughts?". The mistakes are obvious if you actually look. Which makes it clear they didn't. Now you're the one doing the thinking.
• The review that was really the job. Someone asks you to review something technical they generated. You write careful feedback. A few notes in, it becomes obvious they haven't actually read it themselves, and they can't engage with your notes because they don't know the details. The feedback has nowhere to land. Either you take over, or it doesn't get done. The review was never a review. The whole job was always yours.
• The wall of text. A 600-word email where two sentences mattered. They didn't bother to cut it down, so now you have to read the whole thing.
• The drive-by pull request. Five minutes to generate, an hour for some volunteer maintainer to read it, test it, and figure out if it even makes sense.
• The customer support reply. You write in with a specific problem and get back a polite, generated paragraph that addresses none of the details you actually mentioned. The work of being understood just moved to you.
• The "ChatGPT said" citation. Someone tries to settle an argument by pasting in what the model told them, as if that's an authority. It isn't. The model is the start of thinking, not the end of it.
• The hallucinated legal filing. AI writes the brief, invents the citations, and the court ends up doing the fact-checking.
• The unread essay. Assignment generated, never thought about, handed in. The point of doing it was the thinking, and that got skipped.
• The heartfelt note that wasn't. A condolence, a thank-you, a birthday message. Something that was meant to come from a person, and didn't. Even if the recipient can't quite name why, they always know.

The rule

Don't send anyone something you wouldn't be willing to vouch for. Use AI all you like in private. The line is the moment you make it someone else's problem.

Asking for feedback should be the last ten percent of the work, not a sneaky way to offload the first ninety. If you've generated something and you don't really understand it yet, that part's still on you.

This isn't anti-AI

The tools are great. Draft with them, think out loud with them, use them however you want. That's all fine. This is only about the moment you pass something on with your name attached.

If someone you know could do with reading this, send them the link. Or share it on X.

That's it. Thanks for reading.