I think one reason most AI products fail is that we’re too obsessed with the technology.

We start with “How can we use AI here?” instead of “What’s the best way to solve this problem?”

Once you have a model that can do almost anything, everything starts looking like an AI problem.

Sometimes it is.

Sometimes the best solution is a form.

Or a button.

Or a good search box.

Or just removing three steps from an existing workflow.

I saw this while building HealthFirstAI. It would have been easy to keep adding AI because that’s the interesting part. But doctors didn’t need AI everywhere. They needed less typing.

Sometimes the right answer was AI turning speech into structured information.

Sometimes it was just a better workflow.

The technology should fit the problem.

Not the other way around.