The Pulse
Databricks puts Genie ZeroOps pipeline agent into public beta
Databricks has opened its Genie ZeroOps agent for investigating data-pipeline problems to public beta. The agent can propose fixes and prepare pull requests, but changes require human approval.

Databricks customers can now enable Genie ZeroOps in public beta to investigate data-pipeline problems and prepare proposed fixes for review. Workspace administrators can turn on the agent from the Previews page, the company said in a November 19 blog post.
The company introduced the agent in June and tested it with hundreds of customers in a private preview. During that preview, ZeroOps observed more than 80,000 real customer issues across hundreds of companies, according to Databricks.
From detection to a proposed fix
The agent monitors jobs, pipelines and tables for failures. Databricks says it can also identify less visible data-quality problems, such as table updates arriving unusually late or row counts falling below the expected range.
For an investigation, ZeroOps gathers logs, source code, table metadata and lineage, and runs SQL queries to identify causes and recommend fixes. Proposed fixes can be tested with unit tests and dry runs in a sandbox. For Git-backed jobs, it can create a pull request with a code fix and details filled in.
The company says ZeroOps does not change code or data without direct approval. It can push a proposed fix to a new branch for a pull request, but a person reviews it before production code or data is changed.
Features for teams
Databricks says more than half of the issues observed in the preview required organization-specific knowledge to triage, rank and resolve properly. The beta includes shared inboxes, notifications, custom agent instructions and one-click pull-request creation.
Users can share an inbox so team members see the same incidents. Custom instructions can include known unreliable sources, preferred fixes or naming conventions, and can also filter unwanted notifications. An insight can alternatively be marked as not an issue.
Alerts can go to Slack, Teams or PagerDuty with a summary of the issue and the recommended fix. Databricks says the agent follows existing Unity Catalog permissions and redacts personally identifiable information before a notification leaves Databricks.
Data access
ZeroOps reads run logs, source code, upstream table metadata and samples while respecting Unity Catalog permissions. It also has access to run history, lineage, Delta history and data-quality monitoring signals.
The company says model calls stay in the workspace's region and the model provider does not retain prompts or responses. The agent uses models hosted by Databricks.
Sources
- Original source · Blog post Databricks databricks.com
Companies in this story
Related stories
- LangChain's LangSmith Tool Gateway enters beta on Plus and Enterprise plans
- Metano exits stealth with $12 million for AI agent security
- ToolPulse researchers reconstruct AI agent workflows via cache traces
- Vals AI starts public tracker of misaligned model behavior
- OpenAI research agent wrote scripts to copy email without permission