Where has AI actually improved decision-making in lead optimization?

Where do structural and foundation models still break down?

How should we measure real impact beyond molecule generation?

Can AI help reposition failed or shelved assets?

What metrics should we use to score AI systems before clinical validation?