Keep pulling the thread on Matej Zaharia.
Databricks co-founder Matej Zaharia has open-sourced a new project called Omnigent, which is a meta-harness for AI agents.
Merck trained a transformer-based language model called TEDDY on the Databricks platform to predict gene regulatory networks for drug discovery.
The general availability of Iceberg v3 support in Databricks provides a unified data layer, eliminating the need to rewrite data files for sharing between Delta Lake and Iceberg tables.
A unified metadata layer for Delta Lake 5 and Iceberg v4 is expected to be released later in 2026, potentially in Q4.
Databricks is introducing a new query engine called Raiden, which it claims is the world's fastest with latency down to tens of milliseconds.
Databricks is open-sourcing a new project called "open sharing," a superset of Delta Sharing that can share AI assets like agents and models and also works on-premise.
Databricks is launching Unity AI Gateway, a single platform to manage, control, and monitor costs for all AI models and agents within an organization.
Databricks is launching Genie Ontology, a system that automatically constructs a knowledge graph of an organization's data and processes to provide context to AI agents.
Databricks is launching Genie ZeroOps, a background AI agent designed to autonomously monitor, diagnose, and fix data pipelines.
Databricks has agreed to acquire Panther Labs, a SIEM company.
Databricks is launching Customer Lake, an agentic Customer Data Platform (CDP) built on its Lakehouse architecture.
In internal testing, Databricks Genie with the Genie Ontology demonstrated a consistent 30 percentage point improvement in accuracy and roughly half the runtime compared to leading AI agents.