Note · Human-guided AI
AI needs a product direction
The short answer
My thesis is that AI is most useful when someone gives it a clear problem, context and a way to judge the result. It can suggest unexpected combinations and produce alternatives quickly. That does not make every alternative appropriate. Product judgment means choosing whom to help, which tradeoffs matter and what evidence would make us change direction.
Start before the prompt
Imagine an identity check that people abandon. Asking for a more attractive screen skips the important question: are the instructions unclear, is a document unavailable, or does the person distrust the request? Each explanation suggests a different intervention. A polished answer to the wrong question is still the wrong answer.
Give the idea a real test
I would use AI to explore clearer instructions, a recovery path and alternative layouts. Then I would test the experience with people and verify that the underlying system handles failures. Creative exploration matters; so does finding out where the attractive proposal breaks.
Own the whole experience
The same thinking applies to a chatbot or an automated message. What can the person do next? Can they reach a human? Can they stop the messages? My priority is a coherent experience with accountable decisions, not the largest number of generated screens. Speed is valuable when it helps us learn, not just produce more.
Keep reading