The brain dump is uber helpful with almost everything — not just using AI. I always get about 85% through to the solution with the mind map style, arrowed brain dump.
YES, true! Brain dumps are so important and I also think we underestimate them. The name doesn’t help either, I feel 😀 “brain dump” doesn’t sound like the most interesting or useful activity and yet, it’s one of the most valuable ones…
I actually kinda like the name — it's so real, like actually dump it all out. If only people did it more literally, they'll see just how writing all the messy thoughts helps clarify them + lets us connect them better for coherence and direction.
A similar thing happens in those mom/self-help Facebook groups. Either she's got this vague feeling and is actually seeking better language to ask the right question. Or she's just seeking permission to do the thing she's wanting to do in the first place. 🫢 Much could be resolved faster if she took the approach here too.
I actually started lurking (occasionally commenting) in local mom groups, both to see the language (and to steer the responses - not everything needs a parasite cleanse or colloidal silver 👀) and also to use that language for blog posts (as in, “here's the question she should have asked.”).
My version of this is less about taking the wheel back from AI, and more about knowing when to let it keep moving and when to interrupt the frame.
I’ve noticed two useful modes:
“Proceed” keeps the traversal open. It lets the AI expand, test nearby paths, and surface things I may not have seen yet.
“What do you think?” does something different. It asks the AI to compress, evaluate, spot tension, and test whether the structure is actually carrying.
Too much “proceed” and the interaction drifts.
Too much compression and the thinking closes too early.
So the work is regulating that movement: expansion, compression, expansion, compression… without letting the AI’s frame quietly become the boundary of the thought.
Underneath that, I think the deeper rule is:
just because AI can ask, infer, profile, resolve, or complete something doesn’t mean it should automatically become available inside the interaction.
The quality of the collaboration depends on protecting the conditions of the interaction itself.
Not every capability improves coherence.
Not every question deserves an answer.
And not every unresolved tension should be prematurely resolved.
Thanks for sharing this Reuben! I think we're solving for slightly different moments in the interaction though.
What you’re describing feels like it's about the overall flow, but I'm talking about what happens specifically inside AskUserQuestion when Claude fires structured questions at you. Do you use that feature much?
I have never used AskUserQuestion tool but Claude used to question me infinitely anyway. Now I specifically set hard stop on too many questions and limit the recommendations it gives based on my own judgement.
Pushing back against AI is an underrated practice and your penny answer patterns here enforce. Truly writing with your taste, Mia! ✨
Not going to lie, Mia, currently going through a phase where I hate being questioned so much by AI. I have built it into everything, it's needed, it can be good, but ooof. Some days just not in the mood to be interrogated at every step, lol, and I just want to get on with the work using an AI that simply gets it! 😆 I sometimes find AskUserQuestion really good and on the money, other times it's covering ground I have already defined and laid out just before - depends how "lazy" Claude is that day?! 🫣
One thing I learned from a mentor several years ago is that "the quality of your life depends on the quality of your questions". So it's like when someone asks me a question, I take a moment to see if it's not a "lazy question". If it is, I ask myself a better question and then answer that instead. I've carried that to Claude as well.
Sometimes we're in sync and the "recommended" option is what I would respond with. In the others, I choose the "other" or fill in what I believe would be the better direction.
That said, I mostly use Claude on the fields I have expertise in. It would be interesting to see how it would play out in other areas I'm more newbie in.
I think the AskUserQuestion tool fires up when it tries to understand what YOU want it to do. I don't think you can use it for finding out stuff, but now that you brought this up, I'm super curious too 😅
I believe it shows up when it wants to get clarification on plans, approaches, but also when you’re tackling unexplored subjects. In which case I agree with you that it can narrow you down to “normal” spaces. 😆
Mia, this is so good. Annoyingly good, actually. The “Penny Moves” are such a smart fix because they put the user back in charge of the investigation. Especially the brain dump before the AI starts interrogating you with its little clipboard.🩷🦩
So I have a question though about AskUserQuestion. Have you found that it’s getting more concrete (even obtuse!) with the latest versions of the Models (So Claude Opus4.7 and Sonnet 4.6)?
I am refusing to use Opus 4.7, don't like it at all! So I stick with Sonnet and Opus 4.6 :) I feel it works best with Opus. But I haven't necessarily noticed a difference in its behavior from model to model!
Yes, I often use the AskUserQuestion tool a lot, it's really useful in the beginning, since it forces the AI to ask us to help make everything clear when the input or intent is vague. Like you said, it can subtly steer us if we just keep picking from the options, especially the Recommend one. I used to pick it a lot, but recently I often pick "Other" or turn off the box when needed, if I find the question is clearly not related to my goal.
It annoys me A LOT to turn off the box 😅 so I just keep going, and I had to find a few strategies for it hahha. Also, I barely get recommended ones, that's interesting!
The brain dump is uber helpful with almost everything — not just using AI. I always get about 85% through to the solution with the mind map style, arrowed brain dump.
YES, true! Brain dumps are so important and I also think we underestimate them. The name doesn’t help either, I feel 😀 “brain dump” doesn’t sound like the most interesting or useful activity and yet, it’s one of the most valuable ones…
I actually kinda like the name — it's so real, like actually dump it all out. If only people did it more literally, they'll see just how writing all the messy thoughts helps clarify them + lets us connect them better for coherence and direction.
I like your reframe ❤️❤️
You explained the psychological phenomena beautifully. It’s like we switch thinking modes when we use AI. I feel it myself viscerally.
Once you see it, you really can't unsee it! Thank you so much for reading 🥰
A similar thing happens in those mom/self-help Facebook groups. Either she's got this vague feeling and is actually seeking better language to ask the right question. Or she's just seeking permission to do the thing she's wanting to do in the first place. 🫢 Much could be resolved faster if she took the approach here too.
wow, this is a perfect comparison, and the same goes for “medical” facebook groups!
I actually started lurking (occasionally commenting) in local mom groups, both to see the language (and to steer the responses - not everything needs a parasite cleanse or colloidal silver 👀) and also to use that language for blog posts (as in, “here's the question she should have asked.”).
That’s a fantastic strategy!
My version of this is less about taking the wheel back from AI, and more about knowing when to let it keep moving and when to interrupt the frame.
I’ve noticed two useful modes:
“Proceed” keeps the traversal open. It lets the AI expand, test nearby paths, and surface things I may not have seen yet.
“What do you think?” does something different. It asks the AI to compress, evaluate, spot tension, and test whether the structure is actually carrying.
Too much “proceed” and the interaction drifts.
Too much compression and the thinking closes too early.
So the work is regulating that movement: expansion, compression, expansion, compression… without letting the AI’s frame quietly become the boundary of the thought.
Underneath that, I think the deeper rule is:
just because AI can ask, infer, profile, resolve, or complete something doesn’t mean it should automatically become available inside the interaction.
The quality of the collaboration depends on protecting the conditions of the interaction itself.
Not every capability improves coherence.
Not every question deserves an answer.
And not every unresolved tension should be prematurely resolved.
Thanks for sharing this Reuben! I think we're solving for slightly different moments in the interaction though.
What you’re describing feels like it's about the overall flow, but I'm talking about what happens specifically inside AskUserQuestion when Claude fires structured questions at you. Do you use that feature much?
I have never used AskUserQuestion tool but Claude used to question me infinitely anyway. Now I specifically set hard stop on too many questions and limit the recommendations it gives based on my own judgement.
Pushing back against AI is an underrated practice and your penny answer patterns here enforce. Truly writing with your taste, Mia! ✨
Hah, interesting! It never fired up for you randomly? This is sooo weird!
I sometimes feel like the questions go on forever too… haha 😀
Not going to lie, Mia, currently going through a phase where I hate being questioned so much by AI. I have built it into everything, it's needed, it can be good, but ooof. Some days just not in the mood to be interrogated at every step, lol, and I just want to get on with the work using an AI that simply gets it! 😆 I sometimes find AskUserQuestion really good and on the money, other times it's covering ground I have already defined and laid out just before - depends how "lazy" Claude is that day?! 🫣
Love that we approach this very similarly!
TOTALLY relatable! Also, I simply love your back and forths with Claude 🤣
Such an interesting post.
One thing I learned from a mentor several years ago is that "the quality of your life depends on the quality of your questions". So it's like when someone asks me a question, I take a moment to see if it's not a "lazy question". If it is, I ask myself a better question and then answer that instead. I've carried that to Claude as well.
Sometimes we're in sync and the "recommended" option is what I would respond with. In the others, I choose the "other" or fill in what I believe would be the better direction.
That said, I mostly use Claude on the fields I have expertise in. It would be interesting to see how it would play out in other areas I'm more newbie in.
WOW I love that quote! It's perfect!
I think the AskUserQuestion tool fires up when it tries to understand what YOU want it to do. I don't think you can use it for finding out stuff, but now that you brought this up, I'm super curious too 😅
I believe it shows up when it wants to get clarification on plans, approaches, but also when you’re tackling unexplored subjects. In which case I agree with you that it can narrow you down to “normal” spaces. 😆
Mia, this is so good. Annoyingly good, actually. The “Penny Moves” are such a smart fix because they put the user back in charge of the investigation. Especially the brain dump before the AI starts interrogating you with its little clipboard.🩷🦩
Thank you Pinkie! The detective in me is happy 😆❤️
So I have a question though about AskUserQuestion. Have you found that it’s getting more concrete (even obtuse!) with the latest versions of the Models (So Claude Opus4.7 and Sonnet 4.6)?
I am refusing to use Opus 4.7, don't like it at all! So I stick with Sonnet and Opus 4.6 :) I feel it works best with Opus. But I haven't necessarily noticed a difference in its behavior from model to model!
Yes, I often use the AskUserQuestion tool a lot, it's really useful in the beginning, since it forces the AI to ask us to help make everything clear when the input or intent is vague. Like you said, it can subtly steer us if we just keep picking from the options, especially the Recommend one. I used to pick it a lot, but recently I often pick "Other" or turn off the box when needed, if I find the question is clearly not related to my goal.
It annoys me A LOT to turn off the box 😅 so I just keep going, and I had to find a few strategies for it hahha. Also, I barely get recommended ones, that's interesting!
I live for a brain dump !
Any nothing could ever take away your weird, cool thoughts