I’ve found that the best MVP scope is often defined by the decision the product must support, not by the number of features it can ship. I start with one user outcome, identify the riskiest assumption, and design the smallest test that can change our mind. That approach keeps AI work focused and gives the team a clear next step.