This blog post is the last in the series regarding AI skills—particularly the right whys, the why of work.
I asked Claude for concrete, actionable guidance regarding self-leadership.
Here’s part of what it said:
Asking the Right Questions
Lead with generative questions — ones that open territory rather than confirm what you already think. Before any conversation or AI prompt, write the question you want answered, then ask “what would need to be true for this to be the wrong question?” That inversion often surfaces the better one.
To get humans and AI to ask you better questions: share partial thinking, not finished conclusions. Incompleteness invites interrogation. Tell collaborators explicitly: “I want you to poke holes, not validate.”
Thinking Analytically
Frame questions in terms of decisions, not topics — “what should I do about X” sharpens faster than “tell me about X.” Practice steel-manning the opposite of any framing before committing to it.
For finding problems: adopt a pre-mortem habit — assume a plan failed, then work backward. And cultivate discomfort with premature closure; the moment something “feels solved,” treat that as a signal to probe harder.
Critiquing Questions and Answers
For any input (human or AI), ask: What is this question inadvertently ruling out? What assumption is load-bearing here? For outputs: What would falsify this? Who benefits from this framing?
Collectively — in teams or networks — assign someone the explicit role of question auditor: not to answer, but to challenge whether the group is even asking the right things. Rotate that role so it doesn’t become performative.
I asked Claude to elaborate:
On Prompts for Right Questions & Critiquing
Here’s a reusable scaffold for both purposes:
“Before I respond: What assumptions are you making that you haven’t stated? What would make your question the wrong one to ask? What answer would most surprise you — and why aren’t you asking for that instead?”
For critiquing outputs, a companion prompt:
“Don’t evaluate this answer yet. First: what would a strong counter-argument look like? What’s the most important thing this answer might be leaving out? Now critique it.”
The key mechanic: force articulation before AI engages, so the human’s thinking is on record and can be interrogated rather than bypassed.
Pre-Mortem Habit
Coined by psychologist Gary Klein. The discipline: before launching anything, imagine it’s 12 months later and it has failed. Then write down — specifically — why.
Do it in writing, alone first, then share. Group pre-mortems run in reverse order — quiet individual reflection before collective discussion — to prevent anchoring on the first voice.
On Knowing What You Don’t Know
Unknown unknowns have no internal trigger. So you need external structures that fire regardless of awareness. Concretely:
- Scheduled boundary encounters — regularly talk to people in adjacent fields, roles, or worldviews. They will ask questions that feel strange. That strangeness is the signal.
- Audit your question diet — periodically list the questions you’ve been asking. Then ask: what topics, domains, or people are completely absent? Absence is data.
- Use surprise as a sensor — when something surprises you, don’t just update. Ask: what did I implicitly believe that made this surprising? That belief is a candidate unknown unknown.
- Assign a designated challenger — someone whose standing role is to ask “what aren’t we considering?” They don’t need to know the answer; they just need the license to name the gap.
The meta-move: stop trying to know what you don’t know, and instead build structures that make unknown unknowns visible over time — because no single act of reflection will surface them reliably.
A final elaboration request on problem finding:
This requires moving from analysis into creative and anticipatory thinking. Finding problems that don’t yet exist isn’t diagnosis; it’s imagination disciplined by structure.
Finding Non-Existent Problems
You’re not looking for what’s broken — you’re looking for what will matter before anyone else notices it does.
- Trend collision — identify two trends developing independently and ask: when these intersect, what new friction or failure emerges? The problem lives in the collision, not in either trend alone.
- Extrapolate current satisfactions to their breaking point — take something working well today and ask: if this succeeds completely, what does it destroy or crowd out? Success creates its own future problems.
- Borrow problems from adjacent fields — what problems is medicine, architecture, or logistics solving right now that your domain hasn’t encountered yet but will? Problems often migrate across domains with a lag.
- Ask what’s being normalized that shouldn’t be — slow-moving dysfunction becomes invisible. Periodically ask: what are we tolerating and calling it fine? That tolerance is a problem waiting to be named.
- Reverse the beneficiary — ask who is not being served by a current solution, and whether that absence will eventually matter. Unserved populations often define the next problem set.
The animating question across all of these: What will seem obvious in hindsight that nobody is saying out loud right now?