Artificial Intelligence

Databricks says its agent fixes broken pipelines

Databricks moved Genie ZeroOps to public beta after training it on 80,000 pipeline failures.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

Databricks Genie ZeroOps inbox for pipeline incidents
Credit: Databricks

Databricks moved Genie ZeroOps into public beta after testing the agent on more than 80,000 customer pipeline issues, the company said.

Data pipelines break in two ways. A job fails and pages someone, or a table quietly stops updating and no one notices until a dashboard lies.

Databricks found more than half needed company-specific knowledge to fix, which is why off-the-shelf triage keeps failing.

Genie ZeroOps is an autonomous agent that watches jobs, pipelines and tables in the background, then works an incident from alert to proposed pull request without touching production code itself.

It starts with detection of hard failures through job monitoring and silent decay through data quality monitoring, which is Databricks continuous check for late or incomplete table commits. It then pulls logs, source code, table metadata and lineage to find the root cause, tests the fix in a sandbox with unit tests and dry runs, and ranks findings by downstream impact.

It sends a summary with a recommended fix to Slack, Teams or PagerDuty, and for Git-backed Jobs it can open a pull request in one click with code and context pre-filled. The inbox is now shared so a team sees the same incidents, and admins can add custom instructions like ignore this flaky source or follow this naming rule.

Unity Catalog is the Databricks governance layer that controls what data and code the agent is allowed to see, and ZeroOps inherits those permissions while personal data is redacted before alerts leave the platform. That access is the edge over a general coding agent such as Claude Code, which waits for a prompt and lacks built-in run history, lineage, Delta history and quality signals.

ZeroOps sits under Genie, which Microsoft describes in its Azure Databricks docs as the AI experience for asking data questions in natural language, exploring dashboards and running apps.

It still does not push to production on its own. Databricks says ZeroOps never changes code or data without direct approval and only proposes a fix on a new branch.

If you want it, have a workspace admin turn it on from the Previews page, then look for the ZeroOps banner at the top of the pipelines page and add one instruction about your flakiest source first.

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsDatabricksGenie ZeroOpsUnity CatalogSlackPagerDutyGitDelta

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.