join astronomer at Snowflake Summit 2026

Production AI demands the same discipline as a well-run kitchen: everything prepared, organized, and ready before the rush. This talk shows how Apache Airflow unifies data pipelines and AI workflows in one observable control plane — replacing fragmented observability, failure handling, and team handoffs.

Real project walkthroughs include:

• Multi-agent orchestration at production scale
• AI-driven pipeline branching and decision logic
• Vision model integration in live data workflows
• Airflow as a unified system of record for data + AI

Every enterprise has an AI strategy. Few have the foundation to deliver on it. This session tackles one of the most overlooked gaps in production AI — context — and shows the architectural pattern that fixes it. This session will cover:

• Why static context (semantic layers, ontologies) isn't enough for production AI
• How orchestration accumulates the evidence AI actually needs to run reliably
• Capturing what ran, what changed, and what humans corrected — continuously
• The architectural pattern production AI depends on

Financial institutions don't struggle with governance frameworks — they struggle with enforceability. When regulators ask how an AI decision was produced, most teams can't prove which data ran or who approved exceptions.

This session shows how leading institutions use orchestration to make governance real at runtime:

• Codifying governance policy directly into the pipeline — so controls are enforced, not assumed
• AI-driven pipeline branching and decision logic
• Vision model integration in live data workflows
• Airflow as a unified system of record for data + AI