The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks

Sandipan Bhaumik presents a five-pillar framework for successfully moving AI systems from demos to production, inspired by a retail bank's failed chatbot PoC. The framework covers defining numerical success (Evaluation), tracing every AI decision (Observability), building robust data pipelines (Data Foundation), managing multiple AI interactions (Multi-agent Orchestration), and ensuring accountability and security (Governance). He illustrates these concepts with a banking chatbot case study, emphasizing continuous evaluation, data quality, and a proactive incident playbook.