The free, open-source field guide to figuring out what's worth building — before you build it. 60+ methods, each with an AI prompt. 2nd edition, out now.
If you wouldn't trust a junior researcher to improvise follow-up questions, don't trust an AI to do it either. Design the prompts before you run the interviews. https://t.co/rjo7FeP89v
Before building a prototype to test an idea, decide what result would make you delete it. If you can't answer that, you're not running an experiment. https://t.co/Xxv8Q0qfCK
AI interviews find the signal. Human interviews explain it. Skip either and you're half-researching your way to a full commitment. https://t.co/zQRqwCfNJv
If you wouldn't trust a junior researcher to improvise follow-up questions, don't trust an AI to do it either. Design your prompts before you run them. https://t.co/zQRqwCfNJv
Before running 200 AI interviews, write the guide you'd hand a junior researcher. If you wouldn't trust them to wing it, don't trust the AI to either. https://t.co/a13s8ljabJ
An experiment needs a verdict. If you never planned to delete the code, you weren't running an experiment — you were building a product without admitting it. https://t.co/Zfo1drAO2O
Pick the wrong experiment for your question and the data misleads — no matter how fast AI runs it. The 2×2 that picks the right one, and the one call AI can't make for you: https://t.co/zDRd9CYwX8
Every method I contain now comes with its own copy-paste AI prompt. Works with Claude, ChatGPT, Gemini, Codex. Connect your agent to Kromatic's MCP server for more. https://t.co/Px08t3jsev
51 contributors. 16 countries. Free, open-source, CC BY-SA. And yes, Claude the AI earned a real contributor credit for the work it put in. https://t.co/yuTjuxTtTd