Maggie Appleton @maggieappleton.com · Feb 27 Life improved immensely once I learned to stop telling Cursor "it still doesn't work" and instead tell it to pause, don't write any code, investigate the issue properly, come up with 3-5 theories for what's going wrong, and validate theories with logs. I also give it web + docs access, if relevant.
159 likes 11 replies
Replies
I think the real trick here is a mindset shift from "Just fix it for me" to "Let's throughly investigate the problem together and track our active theories about what's happening. Once we validate / invalidate theories, we can consider fixing it" Standard debugging tactics, rediscovered.
I found that Claude Sonnet 3.7 is much better about this. ChatGPT would just keep creating new problems, bit Claude is more careful and methodical.
This is the way!
This is so helpful. Every time I tell ChatGPT it’s wrong and needs to change it doesn’t change
This is eerily similar to how you might coach a junior engineer in this scenario.
This is great, ty. I’ve been using Claude and I’m going to add this to my debugging project instructions. A related strategy is to tell the AI to ask you up to 5 questions about the bug (+ not write code yet). I think both of these also loop you into the problem solving process, which is a win.
When I used Claude to help me figure out a home-brew recipe for Baja Blast, I used a series of prompts designed to prime it with its broadest insights before asking it to provide a recipe. E.g. direct asks entangle too much noise from Gatorade-Sprite based cocktail approximations This works well!
I like that these tools empower total strangers to the field to be able to get something working. But it feels sad when they hit a wall like this and no amount of “still doesn’t work” can fix it and you’re alone. Then prompting skills converge to the problem solving skills that devs use every day.
This works so well!
I've gotten improvements with "run RCA on this issue"
this 👆💯