As a Claude Code user myself. Context is what makes or breaks a bug fix, feature, or great MVP.
I would ask if you’ve found where providing too much detail actually limits the model.
In my experience, sometimes too much context can force the model to be too rigid in its output, creating a bit friction just because it might’ve had a different idea in mind that would have worked better.
But perhaps that’s a programming specific problem
I’ll say I’m still at the beginning of this journey, especially with Claude Code itself. I haven’t yet found that too much information limits the model, but I have found that it’s very easy to manipulate AI to give you the answer you want. So prompting in a way that doesn’t lead the model is crucial, much the same way you want to conduct your user research questions with humans.
Personally, I’ve found feeding a lot of random information without trying to draw connections for AI leads to some very interesting and novel connections. However, this must be grounded in some from of strategy document that makes clear who the user is, what they’re trying to accomplish, and what the business strategy and objective is.
Same here. I purposefully leave gaps to allow it to fill in where I intuitively think it’s able to better than I can.
I think it works better with a touch of purposeful freedom 
Agree on this point. What I believe product and design teams need to understand is that if systems are adapting and morphing in real time, old, static ideas of personas won’t keep pace with the rate at which things are being built.
I don’t think most teams’ thinking has caught up to this gap.