Pinned Tweet
Testing an idea used to cost me a day of hand-coding. Now it's 5 min to prompt, 30 to make tea. That's why I rewrote the book: when building is free, building the wrong thing is the only real cost. kromatic.com/download/the-re…
3
130
Define what you're trying to learn before you write the first line. If I can't name it, I'm not running an experiment — I'm just building. kromatic.com/real-startup-bo…
59
AI lowers the cost of research. It also lowers the cost of fooling yourself. Those aren't separate risks. kromatic.com/real-startup-bo…
76
AI will agree with my startup idea. So will my mom. Neither is market validation. kromatic.com/real-startup-bo…
120
If you wouldn't trust a junior researcher to improvise follow-up questions, don't trust AI to either. Design your prompts before you run the interviews. kromatic.com/real-startup-bo…
81
Talking to 200 customers isn't about 200 answers. It's about finding the 3 questions worth asking deeply. kromatic.com/real-startup-bo…
51
If my experiment only runs on AI-generated responses, I haven't talked to a customer yet. kromatic.com/real-startup-bo…
57
AI lowered the cost of building. Not the cost of building the wrong thing. kromatic.com/real-startup-bo…
1
2
114
Before I prototype an idea, I decide what result would make me delete it. If I can't answer that, I'm not running an experiment. kromatic.com/real-startup-bo…
79
If I can't bring myself to delete the code, I skipped the experiment and started building a product. kromatic.com/real-startup-bo…
74
AI lets me sketch a working product in hours. The hard part isn't building it — it's deleting it when the experiment ends. kromatic.com/real-startup-bo…
1
90
The most valuable output of running AI interviews at scale isn't the answers — it's knowing which questions are worth asking a human next. kromatic.com/real-startup-bo…
1
1
121
A bad AI interview prompt, run at scale, just produces confident noise. The discipline is in the question, not the volume. kromatic.com/real-startup-bo…
64
Speed only matters if you're headed somewhere real. AI helps me move faster — it doesn't tell me if the destination exists. kromatic.com/real-startup-bo…
54
A demo that works is not evidence that customers will buy. It's evidence that the demo works. kromatic.com/real-startup-bo…
1
1
82
The real AI risk isn't wrong answers. It's confidently wrong ones. kromatic.com/real-startup-bo…
71
If I wouldn't trust a junior researcher to improvise follow-up questions, I don't trust AI to either. I design the prompts before I run the interviews. kromatic.com/real-startup-bo…
70
When I can interview dozens of people at once with AI, the bottleneck shifts — it's no longer recruiting. It's knowing what to ask. kromatic.com/real-startup-bo…
71
Running 200 AI interviews isn't about 200 answers. It's about finding the short list of questions worth asking a human next. kromatic.com/real-startup-bo…
98
If I can't delete the code when the experiment ends, I probably stopped treating it as an experiment. kromatic.com/real-startup-bo…
83
AI lets me sketch a working prototype in hours. The hard part isn't building it — it's deleting it once I have my answer. kromatic.com/real-startup-bo…
2
2
121

