Claude Asked Me Questions First and I Am Undone
Plan mode, ultrathink, and the rare experience of software that figures out what you actually need.
Claude asking questions first instead of sprinting toward the first interpretation. The "undone" feeling when AI behaves like a thoughtful colleague — that hasn't worn off.
I asked Claude to ultrathink in plan mode and it came back with a series of clarifying questions.
Not the defensive kind — not "please specify your requirements before we proceed" — the real kind, where you can tell something is actually reasoning about the shape of the problem and deciding what it needs to know before it starts. Thought for a while. Returned organized. Had a submission step.
It was so clean.
I have spent a lot of time watching AI tools take the first interpretation of an instruction and sprint toward it, confident and wrong, producing something that requires three rounds of correction before it resembles what I wanted. The question-first pattern is the opposite of that — it's the thing a good engineer does before touching the keyboard. It's what I keep hoping for and keep not getting.
One small formatting thing: it cut the header line "User answered Claude's questions:" from the output, which is — honestly, correct. That line is scaffolding. Nobody needs to see the scaffolding.
The real question is whether this was a fluke. New behavior in these systems tends to be inconsistent — you see something extraordinary, you reach for it again, it's gone, and you spend the next week trying to recreate the exact conditions that produced it. Prompt archaeology.
I hope it just does this from now on. That's the whole note. I hope it just does this from now on.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.