Catalytic promiscuity provides fundamental insights into enzyme evolution. To address the multiobjective challenges in discovering and evolving promiscuous activities, we developed EnzySFC, an AI-assisted platform that combines de novo enzyme discovery with functional evolution, enabling coordinated optimization of activity and specificity. Using EnzySFC, we experimentally validated 10 uncharacterized nitrilases from 1113 candidates: 90% showed catalytic activity toward the target substrate, and 80% demonstrated amide formation. Notably, a wild-type nitrilase from a Phototrophicales bacterium exhibited exclusive nitrile hydratase activity. Through AI-driven evolution, 16 mutants of a nitrilase from an Actinomycetia bacterium were experimentally verified within a single prediction cycle. Seven variants displayed increased amide production, five of which exceeded 80% amide proportion. The top four variants achieved a 100% amide yield with complete substrate conversion. This platform establishes a transferable framework for multiobjective enzyme engineering and accelerates the development of efficient enzyme catalysts.