Overview
- Databricks announced Genie ZeroOps on Tuesday, June 16, 2026 and said the feature will enter a private preview in the coming weeks with initial support for jobs, pipelines, tables and ML workloads.
- The agent runs a four-step loop — detect, assess, remediate, verify — that uses platform observability to spot failures, Unity Catalog lineage to trace root causes, and agentic code generation to propose fixes.
- Proposed fixes are tested in isolated sandboxes built with zero-copy or shallow clones of production tables so the system can run fixes against real data without touching live production.
- For machine learning, Genie ZeroOps can diagnose silent model drift, train candidate models on corrected features, evaluate them against the production evaluation suite, and help ramp improved models onto live traffic.
- Databricks says users keep control by configuring monitored assets, enforcing Unity Catalog governance, reviewing prioritized incidents in an inbox UI, and approving any change before it reaches production; broad adoption will depend on real-world validation and expanded asset support.