Cisco AgenticOps and Assured Wireless

In this presentation at Mobility Field Day 14, Cisco's engineering and product leadership, including Neil Kulkarni, Benson Lao, and Vishal Desai, detailed the transition from viewing assurance as a feature to treating it as a holistic outcome. The speakers introduced the Closed Loop System framework, which follows a cycle of sensing network anomalies, reasoning through the root cause using AI agents, and acting via either administrative assist or autonomous remediation. By leveraging over two decades of radio resource management (RRM) data, Cisco is up-leveling network operations to move beyond point features toward an Agentic Ops model. This approach is designed to eliminate the manual overhead for administrators, particularly by identifying silent sufferers, users experiencing poor performance who do not file support tickets, and providing contextual recommendations based on the unique signature of each specific site. The technical demonstration featured the AI Packet Analyzer and AI Config Recommendation tools, which utilize a combination of classic machine learning and semantic Large Language Models (LLMs). Vishal Desai explained that approximately one-third of wireless issues require packet-level analysis, a task that historically creates significant mental overhead for engineers. Cisco's solution involves Opportunistic PCAP, where access points automatically trigger packet captures during failures. These captures are then processed by an AI that encodes failure signatures, such as M1/M2 handshake timeouts or certificate validation errors, into human-readable text. This "SME in a box" capability allows even junior administrators to understand complex Layer 1 and Layer 2 issues without manually sifting through thousands of individual trace files. Cisco also addressed the Security and Defense of the AI stack itself, introducing a specialized dashboard for monitoring AI assets, agents, and data training models. This initiative, part of Project Glasswing, focuses on discovering vulnerabilities such as model manipulation and ensuring privacy guardrails for Model Context Protocol (MCP) servers. For configuration optimizations, Cisco is moving away from static golden templates toward site-specific recommendations that use Convolutional Neural Networks (CNN) to analyze long-term and short-term temporal trends. This ensures that changes, such as enabling 802.11r or adjusting transmit power, are cross-validated against experience metrics to guarantee improved roaming latency and capacity. The session concluded by affirming that these tools are being rolled out across both Meraki and Catalyst platforms to provide a consistent assurance outcome for all enterprise customers.