If you use AI to draft messages, make sure you understand the context and ideally rewrite them in your own voice.
When you send raw AI output into chat, design doc, or PR review, you’re more often than not offloading your thinking to the reader. That’s disrespectful of their time but also makes you redundant. If they were fine talking to AI without your input, they’d talk to it themselves.
Current AI models are amazing. They exhibit superhuman knowledge on certain problems and in good hands are an immense efficiency multiplier. They are great at brainstorming and catching problems and are generally the best rubber duck there can be. They should definitely be used.
But they have real flaws. Post-training makes them sycophantic - they’re bad at saying “no” or “I don’t know”. They think too locally, too defensively, and overengineer. Without enough context, they assume the most probable scenario and treat it as confirmed certainty.
Thus, if the output needs to be correct, their conclusions must be carefully validated. The best way to ensure that is to rewrite it in your own voice.
This also lets you add your own expertise - or at least demonstrate full understanding. After all, if all you do is paste LLM output, you’re replaceable by an LLM agent the recipient could run themselves.
When someone realizes you’re pasting AI output, they start wondering: did you even read my message? Do you actually stand behind what “you” wrote? Do you take responsibility for it?
Unedited AI slop signals: “I value my time more than yours.” You saved 2 minutes; the reader now spends 5 parsing fluff.
Inspired by nohello.net and policies of the lovely people at earendil/pi. Source available on GitHub.