Can we trust AI agents in the enterprise? ft. Derek Carr

Is this finally the year AI agents graduate from personal desktop curiosities to mission-critical enterprise systems? While developers have embraced autonomous agents to write code, conduct research, and automate complex workflows on local machines, transitioning those dynamic workloads to production environments introduces an entirely new class of architectural and security dilemmas. Traditional enterprise platforms thrive on deterministic code, GitOps pipelines, and static security policies—assumptions that break down when an agent begins writing and executing code dynamically at runtime.

In this episode of Technically Speaking, Red Hat CTO Chris Wright sits down with Derek Carr, Senior Distinguished Engineer at Red Hat and longtime Kubernetes pioneer, to demystify the internal anatomy of AI agents and chart the path to enterprise readiness. Using a tangible physical analogy—mapping the LLM to the brain, tooling to the hands, context to short- and long-term memory, and harnesses to the nervous system—Derek breaks down why treating an agent as just another containerized workload falls short. Moving an agent from a laptop into a centralized cluster exposes critical questions around enterprise identity, state retention, multi-user concurrency, and credential exposure.

The conversation explores how open source communities are answering these challenges by establishing strict trust boundaries and sandboxing execution environments. Derek details the emergence of OpenShell—an agent execution runtime built to decouple sensitive credentials from untrusted actions through granular, deny-by-default runtime policies. Chris and Derek also examine the co-evolution of projects like OpenClaw and the newly announced OpenClaw Enterprise architecture, demonstrating how collaborative open source governance is turning early viral experiments into hardened, auditable, and multi-tenant enterprise platforms.

Tune in to discover how sandboxing, runtime policy enforcement, and defense-in-depth architecture are laying the groundwork to safely unlock autonomous AI agents across the enterprise.