The industry is captivated by the promise of AI agents, but most of the attention has gone to capabilities, not controls. Agents are designed to be persistent, goal‑seeking systems that will iterate, retry, and escalate until they achieve an outcome. Without guardrails, that same tenacity can lead to runaway token consumption, a denial‑of‑service‑level demand on data systems, and malware‑like probing of sensitive or privileged resources.

This raises the real questions enterprises are wrestling with: Why do autonomy controls remain so immature — whether identity‑based or the contextual, temporal, geographic, and workload‑aware governance that agentic AI will ultimately depend on?

And beyond autonomy: Where are the mechanisms for testing, versioning, controlled deployment, and rollback? For all the talk about the "agentic enterprise" and emerging agent control planes, why does operational rigor still seem to be missing?

This panel brings together experts who are confronting these issues directly. We’ll examine what governance for agentic AI must look like, what’s missing today, and what enterprises need before agents can be trusted in production. Most important, we want to hear your concerns, objections, and real‑world constraints -- because those are exactly the signals the industry needs to get this right.