Natural product discovery is now bottlenecked by connecting complex data instead of acquiring it. We show how integrating protein language models and mass spectrometry workflows can prioritise microbial producers of valuable natural products beyond standard reference-driven approaches.