AI Agent News/stories/databricks-opens-public-beta-of-genie-zeroops-pipeline-fixing-agent.md

Infrastructure

Databricks opens public beta of Genie ZeroOps pipeline-fixing agent

Databricks has moved Genie ZeroOps, an agent that detects, investigates and proposes fixes for data pipeline issues, into public beta. Workspace admins can switch it on from the Previews page, and the product draws on observations from more than 80,000 customer pipeline issues.

Source: www.databricks.com/blog/genie-zeroops-now-public-beta-autonomous-agent-fixes-data-pipelines-lessons-80000-customer

Genie ZeroOps watches jobs, pipelines and tables for hard failures as well as silent freshness and completeness problems. It collects context from logs, source code, table metadata and lineage, runs SQL queries to identify root causes, and tests proposed fixes with unit tests and dry runs in a sandbox.

The beta adds a shared inbox, notifications with issue summaries sent to Slack, Teams or PagerDuty, custom instructions for team-specific rules, and one-click pull request creation for Git-backed Jobs. Databricks says over half of the issues it saw during private preview needed organization-specific knowledge to triage. ZeroOps follows existing Unity Catalog permissions, and PII is redacted before notifications leave Databricks.

Databricks says ZeroOps will not change code or data without a person's approval. Its models are hosted by Databricks inside its security perimeter, and the model provider does not retain prompts or responses. Users can filter expected failures with a custom instruction or mark an insight as not an issue.

Why it matters if you build agents

Teams building agents that touch data pipelines can use a managed triage agent rather than maintaining an internal prototype, which Databricks says becomes hard to operate across many pipelines. Access is limited to what Unity Catalog already allows, so permission design carries over.

Fixes still go through a pull request that a person reviews, so builders should plan for that review step in any automated remediation flow. Silent failures depend on Data Quality Monitoring signals, so freshness and row-count checks need to be configured for them to be caught.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe