In practice
It includes customer conversations, behavior research and cheap tests of an idea. The goal is to lower the cost of being wrong. A conversation costs an hour. A feature nobody needs costs a build and then stays in the codebase.
Discovery isn't market research, which asks whether a market exists. It asks what a specific group is struggling with in enough detail to guide a build.
Why it matters for product builders
Most wasted engineering work isn't badly built. It is well built for a problem nobody had.
What changes in AI-first products
Nothing about customer needs got faster. Discovery matters more because building is no longer the main constraint.