Over the past year, I've shipped a handful of AI features — some flopped, a few took off, and a couple genuinely surprised me. Building in public has been the best way to learn what actually makes an AI product useful, because real users have a way of telling you what matters.
In this essay, I want to share the lessons that stuck. These aren't abstract theories — they're practical takeaways from building, testing, and iterating on real products with real people.
Usefulness is contextual
An AI feature isn't useful in the abstract. It's useful in a specific situation, for a specific person, at a specific moment. The same model can feel magical in one context and completely irrelevant in another.
“Usefulness isn't about what AI can do. It's about what it helps people do.”
The best feedback I've received always references a real job, task, or moment. Start with the human context, not the technology.
Speed changes behavior
AI's speed isn't just a nice-to-have — it changes how people work. When something takes seconds instead of minutes, people try it more, iterate more, and think bigger.
The faster the feedback loop, the more likely people are to make it part of their routine.
Real workflows beat demos
Demos are great for showing what's possible. But real workflows are what create lasting value. People don't want to try AI — they want to get something done.
Distribution matters too
A useful product that no one can find isn't useful. Distribution is part of the product. Meet people where they already are and make it easy to discover, easy to try, and easy to share.
