When an AI agent can query sensitive enterprise data via natural language, 'governance'...

When an AI agent can query sensitive enterprise data via natural language, "governance" isn’t just a policy—it’s the backbone of your architecture.

I recently finished listening to a fascinating discussion on the AI in Infrastructure podcast with Anahita Tafvizi, Chief Data & AI Officer at Snowflake, and Daniela Faggella.

They broke down the critical necessity of robust, role-based access control (RBAC) in this new era of agentic AI.

For those of us building for production-grade AI, a few key takeaways stood out:

• Standardization is Key: To move from a system of record to a system of action, we need repeatable, standardized agentic workflows—not just fragmented pilots.

• RBAC is Non-Negotiable: In regulated industries like finance, HR, and sales, an agent’s access must be as granular as a human’s, if not more so.

• Embedded Governance: Governance can’t be an afterthought; it must be embedded by design to ensure the agents we deploy are trustworthy and secure.

Practical Implementation: The discussion highlights the vital work of securing new points of entry for AI, the importance of standardizing metrics across the organization, and how embedded data teams are essential in designing trustworthy, agentic systems from the ground up.

How is your organization rethinking data access control as you scale agentic workflows?

Would love to hear your thoughts in the comments.