A new Ask Good Questions pair of articles is up.
This one is about the little phrase that can quietly get AI coding sessions into trouble:
“And one more thing.”
When we’re working with AI, it’s easy to keep piling extra tasks onto the same thread because the AI is fast and already “in the code.” But in a larger project, that can lead to local fixes that solve the visible problem while disturbing something else in the system.
The guide covers when task stacking is okay, when it needs a fresh thread, and why “enough room for words” isn’t always enough room for judgment.
Guide:
https://askgoodquestions.dev/guides/and-one-more-thing
Field note:
https://askgoodquestions.dev/field-notes/the-day-one-more-thing-became-the-real-bug
Charles
Ahh that. I think I mentioned this, but I use one prompt (sometimes a different model) to define my goal and prompts efficiently before send them to the real coding agent. The prompter as I call it, sometimes tells me “do not derail of the X thing, that’s a different task”.
As you said, make us feel like we have super powers and we entangle the tasks mixing things up while working.
Exactly. That “prompter” step is a really good habit.
I agree that working with AI really does make us feel like we have superpowers at times. But then the downside is that we start stacking three or four “while you’re in there…” ideas onto the same task.
Unfortunately about half the time the AI is happy to follow us right off the cliff.
That little warning: “don’t derail into X, that’s a different task” is exactly the kind of guardrail I’m talking about.
Sometimes the best prompt is not the one that tells the AI what else it could do.
It’s the one that reminds both the AI and the human what we agreed we were actually fixing.