AI widens your options, you do the choosing
You start from a sourced opportunity, write your own ideas before the first prompt, get two models to diverge at volume, then narrow down on a written grid. You leave with five options that survived an attack, each with a testable hypothesis.
Why
Ten ideas asked from an AI are ten average ideas that look alike. A model's ideas are good on average, and more similar to each other than a human group's. Show writers a single generated idea and their stories drift closer together: the least inventive gain, the most inventive gain nothing, and the group as a whole narrows. This kit keeps what AI does better than you, volume, and takes away what it does worse: novelty, sorting, proof. On synthetic users my position is firm: hypotheses yes, validation no. An option is not validated because a model likes it. It is validated when a real user…
When
You build this kit between an established opportunity and the first screen. Phase: Discovery. Audience: product designers, PMs and POs who run ideation. ✅ You know it's your kit when You have a sourced problem and no option yet Your last workshop ended with dozens of ideas and no decision Your ideas look like the AI's ever since you let it speak first ❌ Not for this kit You have no opportunity…
Tools included
- Claude
- Gemini
- Figma