Audit an AI-built BigQuery pipeline with Claude Code

Watch a full data-quality audit of a BigQuery pipeline that ran green while writing wrong numbers.

Who it's for: Data engineers and analysts running a pipeline an AI agent helped build.

The problem: A pipeline can finish every run without an error and still write wrong numbers. One day reported 1,589 views when the real figure was 566.

Claude Code and Codex audit the warehouse data instead of the code, prove each finding with a query, then back up, repair and verify the table.

Takeaways:

  • Point the agent at your data, not only your repo.
  • Make it prove every finding.
  • Back up a table before any repair.
  • Reuse the audit prompt from the repo.

Resources