What is the most important oversight for agentic AI deployment?

JA

James J Azar

Chief Information Security Officer · Artera Services

Least-Privilege Access

James J Azar says the guidance emphasizes least-privilege access to tools and data, with agents receiving only the permissions necessary to perform their defined functions.

Rama Akkiraju

VP, Enterprise AI · Nvidia

Observability

Rama Akkiraju says observability is key, especially to understand agent behavior before certifying an agent.

AH

Anders Hejlsberg

Technical Fellow · Microsoft

Vouch for Responsibility

Anders Hejlsberg says the practice is to understand what you're vouching for before giving AI the keys, because someone must take responsibility and he wouldn't sleep otherwise.

Rashmi Shetty

Senior Director - Enterprise Generative AI Platform · Capital One

Evaluate the Whole System

Rashmi Shetty says the key oversight practice is evaluating the whole agentic system, not just the LLM model alone, because evaluations provide the evidence package of whether the system accomplished its intended task.

The answer starts at 14:53.

Kumar Veetrag

Chief Technology Officer · Enparadigm

Continuous Oversight

Kumar Veetrag says continuous oversight, along with human judgment and ethical governance, is required rather than treating AI as a fully autonomous decision maker, because building trust in AI requires it.

Mandy Andress

CISO · Elastic

Inventory AI Usage

Mandy Andress says the first priority is getting an inventory of AI usage, especially agentic AI, because if you don't know where it's used and its objectives, it's harder to put controls around it.

The answer starts at 15:01.

Konstantin Klyagin

Founder · Redwerk

Test Contradictory Inputs

Konstantin Klyagin says the most important part of testing agentic AI is trying contradictory inputs rather than just the happy path, because people are not always consistent.