Does the AI agent fix pipelines without human approval?

No. The agent proposes the repair and opens a clean pull request with full context, but a person still reviews and decides what ships.

What results has Snowflake seen from the agent so far?

Over five months on the pipelines its team owns, Snowflake reports near-complete triage coverage, under 16 minutes from detection to PR, hundreds of PRs merged, and 500+ pipelines improved.

How are the agent's workflows configured?

Workflows are defined in JSON, specifying the instructions file, accessible repos, a turn budget, which model to run, and where to send notifications.

Large Language ModelsAI Business & IndustrySnowflake AI Blog2d agoOct 20, 2026, 04:00 JST

Snowflake AI agent fixes pipeline failures in under 16 minutes

Send on LINE

3 Key Points

What happened
Snowflake's internal data team deployed an agent that investigates every monitored pipeline failure and opens a ready-to-review PR in under 16 minutes on average, yielding hundreds of PRs merged and 500+ pipelines improved in five months.
Why it matters
The agent also saved hundreds of daily wasted credits by surfacing compute waste from underperforming pipelines that teams rarely inspect, and it inspired Snowflake's product team's autonomous data engineering plans.

WHO IT HITSData engineering and platform teams running large pipeline fleets stand to benefit most, since the agent triages failures and drafts fixes for human review rather than replacing engineers. Snowflake customers may eventually see similar capabilities, as the internal work directly inspired the product team's plans.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

The build started from a familiar pain point: pipeline counts grew faster than the team could, so failure volume rose even though individual pipelines weren't getting worse. Each new pipeline added another place to fail and another dependency that could break at 2 a.m.

Snowflake's team initially considered a Snowpark Container Services server running a REST API, but that approach capped how many runs a single server could handle. The chosen architecture allows many agents to run in parallel without that bottleneck, and the team plans to move to EXECUTE AGENT TASK from CoCo automations, now in public preview, where one SQL call triggers an agent run and Snowflake handles the rest.

The agent's scope runs the full cycle: scanning for code smells and performance issues, catching failures proactively, tracing root causes through logs and the codebase, testing fixes before production, and opening a PR with full context. Snowflake says the internal work directly inspired what its product team plans to bring to customers, though it notes forward-looking statements are not commitments to deliver product offerings.

FAQ

Does the AI agent fix pipelines without human approval?: No. The agent proposes the repair and opens a clean pull request with full context, but a person still reviews and decides what ships.
What results has Snowflake seen from the agent so far?: Over five months on the pipelines its team owns, Snowflake reports near-complete triage coverage, under 16 minutes from detection to PR, hundreds of PRs merged, and 500+ pipelines improved.
How are the agent's workflows configured?: Workflows are defined in JSON, specifying the instructions file, accessible repos, a turn budget, which model to run, and where to send notifications.

Snowflake AI BlogRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.