Imagine being in a hospital exam room, witnessing a medical transcription agent updating electronic health records, suggesting prescription options, and displaying patient history in real time. At the same time, picture a computer vision agent on a manufacturing line conducting quality control at speeds beyond human capability. These non-human identities generated by AI agents present a challenge for enterprises in terms of inventorying, scoping, and revoking identities at machine speed.
The issue holding back agentic AI from moving beyond pilot stages is not the model capability or computational power but rather identity governance.
Cisco President Jeetu Patel highlighted that while 85% of enterprises are running agent pilots, only 5% have transitioned to production. This gap reflects a trust problem, with key concerns revolving around which agents have access to sensitive systems and who is responsible when an agent exceeds its designated scope. Research has shown that many businesses lack role-based access control robust enough for managing human identities, let alone AI agents.
Michael Dickman, SVP and GM of Cisco’s Campus Networking business, emphasized the importance of a trust framework that focuses on actual system-to-system communications rather than inferred activity. This approach enables organizations to enforce agent policies effectively at machine speed.
Dickman outlined four key conditions to address the trust gap posed by agentic AI: secure delegation, cultural readiness, token economics, and human judgment. Establishing secure delegation involves defining an agent’s permissions and maintaining human accountability, while cultural readiness entails adapting workflows to accommodate agent-scale processing.
Furthermore, the network’s ability to capture real data communications offers insights that traditional observability tools may miss. By unifying network, security, and application telemetry, organizations can enhance cross-domain visibility and correlation, essential for deploying production-ready AI agents.
Addressing the trust gap requires actions such as auditing agent identities, implementing microsegmentation for blast radius containment, and establishing a governance-to-enforcement pipeline for seamless policy enforcement. By aligning cross-functional teams, enhancing identity and access management governance, adopting platform-based networking infrastructure, designing hybrid architectures, and prioritizing high-value use cases for trustworthiness, organizations can accelerate the deployment of AI agents.
Ultimately, the organizations that prioritize identity governance, cross-domain visibility, and policy enforcement will lead the way in deploying AI agents successfully. Trust is the foundation that enables the rapid deployment of AI technologies, and organizations that establish robust trust architectures will outpace their competitors in the AI landscape.
