Data Science Agent runs BigQuery ML from features to eval

The walkthrough illustrates how BigQuery's Data Science Agent fits alongside other agents in BigQuery Studio—Data Engineering Agent for pipelines, Data Preparation Agent for cleansing and Conversational Analytics for exploration—so users can route each task to the appropriate helper. It also highlights practical cautions: generated SQL must be checked for data leakage between features and the target, the default 0.5 threshold should be adjusted to favor recall or precision depending on the campaign, and training costs scale with data volume, so asking the agent for a dry-run estimate beforehand helps with cost control. The agent's ability to produce a v2 model with added features and compare metrics against v1 suggests an iterative improvement loop driven by natural-language prompts.