Databricks OpenSharing: AI Agent Skills Sharing Solved

Databricks has launched OpenSharing, an updated protocol designed to improve how AI agent skills are shared. This new version builds on the open-source Delta Sharing protocol. OpenSharing now supports Apache Iceberg REST Catalog clients and includes on-premises storage vendors like Everpure, MinIO, and Qumulo. It's also a standalone Linux Foundation project. The protocol extends zero-copy sharing beyond tables to include agent skills, AI models, and unstructured data. Akram Chetibi, from Databricks, states that agent skills are the most requested new asset type. He highlights the current difficulty of sharing agent skills, often relying on emailing files, which creates update challenges. Databricks calls OpenSharing the first open, vendor-neutral protocol for securely sharing AI assets. This matters because it aims to simplify and standardize the sharing of AI capabilities between organizations.