An idea can sound promising because it connects things we already care about. AI and climate is an obvious example: powerful new tools, important problems, and a great deal of room for imagination.
But a connection between two interesting fields is a starting point. It is not yet a reason for someone to use a product.
The question I want to ask earlier is: whose task would change? If I cannot describe the person, their current workaround, and the decision or action they need to take, the idea is still too abstract to test well.
Inversion adds another question: how could this experiment produce an encouraging result while teaching us very little? We could invite only friends. We could demonstrate a polished interface without checking the underlying answer. We could measure enthusiasm when the real question is repeated use.
Working backward from those failures helps make the test more useful.
A useful experiment
Write a one-page test brief: user, current task, proposed improvement, weakest assumption, and a result that would make you stop or change direction. Keep the test small enough that learning does not depend on a large launch.
Question still open
What is the smallest real-world test that would distinguish an attractive climate-tech idea from an improvement someone actually needs?
Author note
This is a thematic draft proposed for the September retrospective. Add a specific experiment and its evidence before publishing.