Convergent vs. Discriminant Validity: Which Is More Important?

I agree that convergent validity is more important, because if items within a construct are not related, the construct is not measured properly.
In my research, I also focused on factor loadings and AVE first to ensure consistency before analyzing relationships. Without this, the model would not be meaningful.
However, discriminant validity is still important later to make sure constructs are distinct.