DataFramed #328 The Challenges of Enterprise Agentic AI with Manasi Vartak, Chief AI Architect at Cloudera
109 snips
Oct 27, 2025 Manasi Vartak, Chief AI Architect at Cloudera, discusses the complex landscape of enterprise AI. She delves into the importance of having a robust governance framework to manage privacy and security challenges. The conversation highlights the necessity of understanding data lineage for accountability and the distinction between using private AI and cloud services. Vartak also addresses the skills needed for teams to adapt, the balance between prototyping and production, and exciting future directions for agentic AI in various industries.
AI Snips
Chapters
Transcript
Episode notes
Pick Deployment By Scale And Regs
- Choose deployment based on scale and regulatory needs; small scale favors cloud, large scale can justify on-prem costs.
- Model inference often becomes cheaper on-prem once you run many models or face regulatory constraints.
Benchmark On Your Use Case
- Evaluate models on your own use cases instead of trusting public benchmarks that can be gamed.
- Run apples-to-apples comparisons with your real data and tasks to choose the best model.
Grow AI Evals Iteratively
- Start AI evaluations with a small set of test cases (10–20) and expand them organically as failures appear.
- Build evals that assert desired properties, not just exact outputs, to catch non-deterministic mistakes.