Why App-Level Permissions Fail for AI Agents: Okta & NanoCo at Disrupt 2026

As enterprises rush to deploy agentic AI, they're discovering that traditional application-level permissions are insufficient for securing autonomous agents. In this Disrupt 2026 session, Okta's President of Products and Technology Ric Smith and NanoCo Co-Founder and CEO Gavriel Cohen explain why agent security requires a fundamental rethinking at the infrastructure level. They discuss the architectural choices that matter most when building secure foundations for AI agents, from identity management to permission scoping. The conversation highlights how companies are rebuilding cybersecurity basics from scratch to accommodate agentic AI, which operates differently from human users or traditional software. Smith and Cohen share insights on what it takes to secure AI agents without relying solely on application-level controls, emphasizing the need for infrastructure-level safeguards. They also explore the trade-offs and design decisions that engineering teams face when integrating agents into enterprise environments. This session is essential viewing for security architects, AI engineers, and technology leaders planning to deploy agentic AI at scale. Learn from two industry experts about the pitfalls of conventional permission models and the infrastructure patterns that enable safe, scalable agent deployments. #TechCrunchDisrupt2026