12 Questions Every Founder Should Ask

A model that is right 95% of the time and trusted 20% of the time is a design problem, and your first AI designer is the hire who solves it. These 12 questions separate the shippers from the theorists — and I sit the exam myself, in writing, at the technical screen.

Questions to Ask

1. “Tell me about a time you shipped an AI product and it failed. What went wrong, and what did you learn?”
   Look for: Specific story (not generic), acknowledgment of user pain, concrete lesson learned. Avoid: “It worked perfectly” or vague answers. Failure teaches; success is often luck.
2. “What’s your approach when the AI model is right 95% of the time but users only trust it 20% of the time?”
   Look for: Understanding of human-in-the-loop design, explainability, confidence scoring, failure state design. Avoid: “Just show better metrics” or “Better marketing.” This is a design problem, not an algorithm problem.
3. “Can you write HTML, CSS, and JavaScript? Or do you only design in Figma?”
   Look for: Yes to code. Designers who code can prototype ideas in hours, not weeks. They understand constraints. They ship faster.
4. “How do you think about edge cases in AI design? Give me an example.”
   Look for: Thinking about failure states, empty states, what happens when confidence is low, distribution shift, user error. Avoid: Only talking about happy paths. Edge cases are where trust breaks.
5. “Tell me about a designer hire you’d recommend or not recommend.”
   Look for: Clarity on what makes someone good (shipper vs. perfectionist, systems thinker vs. pixel-pusher). Avoid: Vague praise. Good founders have opinions about talent.
6. “Walk me through how you’d design the confirmation flow for a 98% accurate recommendation engine.”
   Look for: understanding that asking for confirmation 98% of the time is broken UX — a score should resolve to a verb (act, review, or ignore), with undo patterns and override paths. Avoid: “Just add a confirm button.”
7. “What’s the difference between AI that helps users decide and AI that replaces human judgment?”
   Look for: Nuanced view of human-in-the-loop, understanding that different domains have different rules. Avoid: “Just automate it.” Judgment-heavy domains (legal, finance, medical) need different patterns than simple classification.
8. “How do you test design decisions with real users instead of just stakeholders?”
   Look for: User testing, design research, willingness to be wrong. Avoid: “I just know what users want.” You don’t. Users do.
9. “What’s the biggest mistake you see founders make when building AI products?”
   Look for: Thoughtful critique (overoptimising the model instead of human-in-the-loop design, ignoring edge cases, poor failure state design). This shows they’ve seen patterns.
10. “How do you stay current with new AI/design trends?”
    Look for: Reading, thinking, shipping. Avoid: “I use ChatGPT” (everyone does). Real: “I shipped X and learned Y, which is why I now think about Z.”
11. “Tell me about your worst hire decision. Why did it go wrong?”
    Look for: Honesty, learning. Avoid: “I’ve never made a bad hire.” Everyone has. Bad hires teach more than good ones.
12. “What’s your definition of shipping? How do you know when something is ‘done’?”
    Look for: Shipped products, clear definition (users > perfection), comfort with iteration. Avoid: “When it’s perfect” or “I don’t know.” Shipping is the skill. Perfection is a failure mode.

Red Flags

• “I only do X tool.” ← Tools change; thinking doesn’t. Flexibility matters.
• “I wait for perfect specs before designing.” ← Shipping requires iteration. You’ll never have perfect specs.
• “Users don’t know what they want.” ← True, but they know what hurts. Start there.
• No actual shipped products. ← Potential is not proof. Where’s the work?