Ask HN: Have you used LLMs in software design/planning?
I realize this may come off as naive, so I’d like to start by saying that LLMs are extremely error-prone. I wouldn’t trust them to create a plan.
That said, I’m wondering if they couldn’t be used in a more loose way to suggest things. For example, if you’re planning on building out a feature, maybe it could suggest potential failure modes that you might take into account. In the best case, it might suggest a pitfall you wouldn’t have seen coming. Of course it’s still your job as an engineer to evaluate whether those concerns should be taken seriously.
It's more helpful if you give it the plans you already have and then ask how you'd improve them.
Less helpful as a starting point IMHO.
I’ve used Claude to ask me questions about my plans and potential features. It’s pretty good for that. Letting it make the plan itself? Nah.