AI doesn't make decisions better than humans — but it makes human reasoning more structured, surfaces information faster, and catches blind spots before you commit.
The mistake: asking AI for the answer. The right approach: using AI to make sure you're asking the right question.
02 Weak vs. Strong
EX 01Build vs buy — auth system
We're making a build-vs-buy decision for authentication at Vitae (8-person startup, pre-launch B2B SaaS).
Option A: Build custom auth (email/password + magic links + SAML for enterprise)
Option B: Auth0
Option C: Clerk (newer, developer-focused, built-in UI components)
Constraints:
- 2 Go engineers, neither has built auth before
- Need SAML within 6 months (required for our first $50k enterprise deal)
- Budget: ~$500/month max on infrastructure
- Timeline: basic auth working in 3 weeks
- Scale: 500 users launch, 50k in 2 years
Structure this decision for me:
1. The real tradeoffs (not surface pros/cons): what are we actually trading off beyond 'time vs control'?
2. Asymmetric risks: which option has the worst worst-case? How likely for a team like ours?
3. Reversibility: if we choose Option B, how painful is migrating in 18 months?
4. Hidden costs most teams underestimate for each option
5. Decision criteria: 2–3 questions whose answers would determine the right choice for our specific situation
Do not give me a recommendation — help me think through it better.
→ Why it works
'Do not give a recommendation' keeps human judgment in control. Five thinking frames each address a real blind spot.
EX 02Hiring decision — two final candidates
Deciding between two final-round candidates for Senior Backend Engineer at Vitae.
Candidate A: Yuki, 9 years exp, Go + PostgreSQL expert, ex-Stripe/Cloudflare, asking $190k. Methodical, thorough, takes time before answering. Some hesitation on ambiguous technical problems.
Candidate B: Marcus, 6 years exp, strong Python/Go, 2 startups (one failed, one acquired for $30M), asking $160k. Moves fast, opinionated, pushed back on one architecture decision in the interview (respectfully but clearly).
Help me think through this:
1. What I'm probably underweighting: given only what I've told you, what factor am I most likely underweighting that matters for an 8-person startup vs. a large company?
2. The regret test: what does the failure story look like for each hire at our stage?
3. The question I haven't asked: one interview question for each that would directly resolve my biggest uncertainty about them
4. What the choice reveals: if I hire Yuki, what does that say about the company we're building? If Marcus?
Do not tell me who to hire. Help me decide better.
→ Why it works
'Do not tell me who to hire' keeps judgment where it belongs.
03 Key Points
01AI generates options you haven't considered — unconstrained by your habits
02Ask AI to argue against your plan before you commit
03'What would I need to believe for Option B to be better?' exposes assumptions
04Generate a decision memo with AI — writing it reveals gaps
05Never outsource irreversible decisions — AI for structure, humans for judgment
04 Model-Specific Notes
Claude is an excellent thinking partner for complex decisions — it pushes back, identifies assumptions, and generates genuine counterarguments. Opus extended thinking is worth activating for high-stakes decisions.
05 For Your Role
Before any important decision: 'What are the 3 most important things I should consider that I'm probably NOT thinking about?' Consistently adds value.