Over the last few years, VRT has been investigating possible use cases of AI in the music domain. Having done some POCs with diverse companies, some lessons are learnt, others are still unclear. The expectations in, and purposes of, AI are very broad and diverse: Reducing workload in cataloguing large collections, enriching them with subjective metadata such as moods, removing the subjectivity when classifying the music in genres, and facilitating music selection for the public are among them. AI also comes with specific challenges regarding the need for good editing tools and good search interfaces, right thresholds for appropriate search results and usable taxonomies and ontologies. In this presentation, VRT shares some insights regarding the POCs that have taken place and discuss next steps and priorities for the future.