A few days ago I wrote here that customers shouldn't have to choose between moving fast and moving safely. Last week we published new research with Omdia that puts some numbers behind why. The average organization now generates about 4,100 monitoring alerts a day, and about half of them are network-related. In that same day, a single practitioner can work through about 21 of those alerts. The result is predictable: 46% of network alerts are closed without investigation. You cant hire your way out of this one. The leaders in this study know it. More than half are already using agentic AI that acts on NetOps processes and tasks, and 84% expect an AI-led operating model within twelve months. Trust is also key: 99% of leaders wont let AI act without guardrails. Nobody I talk to wants less AI in their operations. They want AI they can see, explain, and control, and that has to be built into the infrastructure itself. The network is foundational to both AI’s progress and its governance. Thats the standard we're building to with Cisco AgenticOps: agents that act, with people still in command. Thank you to Joe Vaccaro and the team, and to Jim Frey and our partners at Omdia, for the work behind this research! The full report is here, and worth a read: https://cs.co/6006BG97LY
This is where I think the conversation around AI gets really interesting. Moving faster is great, but people still need to understand what’s happening and feel confident stepping in when needed. Keeping people in command while letting technology take on more of the workload feels like the right balance. Great perspective, Chuck.
Cisco’s AgenticOps approach captures that vision perfectly, empowering practitioners to act faster, smarter, and more securely while maintaining full visibility and control. This kind of innovation defines the future of resilient digital ecosystems. Kudos to the Cisco and Omdia teams for quantifying what many of us experience daily. The numbers tell a compelling story, and the solution points to a new era of AI‑driven operational excellence.
There’s a layer above trust, transparency, and unified interfaces that determines whether agentic systems are actually safe: *architectural stability*. Agentic AI doesn’t become risky only when it makes a wrong decision; it becomes risky when its *internal state transitions* drift. Latent‑space discontinuities, envelope‑boundary deformation, and agentic‑escalation thresholds occur before any alert, remediation loop, or “blast radius” becomes visible downstream. Some of today’s alert flood and multi‑domain anomalies may not be “noise” or “complexity,” but symptoms of upstream instability in the agentic substrate itself. Transparency shows *what* the agent did. Unified interfaces show *where*. Guardrails show *whether* it was allowed. 👉 Architectural stability shows whether the system was *ever safe enough to act*—and ultimately matters more than governing output behavior. 👈 ⚡ The substrate is assumed stable. It is not. It is *dynamic* and constantly changing due to being acted upon. The Stability Envelope: https://catchproof.square.site/the-stability-envelope As AgenticOps becomes the operational backbone, governing the physics of the system will determine whether autonomy scales safely… or simply accelerates instability.
There is no doubt in my mind that the topic of how we all innovate safely and at pace in an age of AI is going to dominate the board and c-suite aganda of Enterprises for the foreseeable....And we were delighted when we were invited to support Cisco in the development of their thought leadership. From Jeetu Patel’s visionary keynote at Cisco Live US 2025 to the exciting product launches under Cisco Cloud Control this year. Kudos to all.
The alert math makes a strong case for why the operating model has to change. What gets more important as agents move from recommending to acting is defining what people still being in command actually means. Visibility and explainability matter, but so do clear authority boundaries, approval rules, override rights and accountability for the actions agents are allowed to take. The more autonomy we give the system, the more deliberately we have to design the human control around it.
Chuck Robbins highlights this exact breakdown: 4,100 daily alerts vs. 21 handled per person. Closing 46% uninvestigated proves we can't solve exponential noise with linear headcount.
Great insights on Ai and how to develop customer friendly and assuring infra where they can easily and eyes closed rely on this machines operating in the background one thing for sure i always think that we don’t need traditional firewall behind ai we need dedicated firewall which can recognise what ai is capable of and act accordingly.
The math has officially broken, human-scale triage cannot govern machine-scale telemetry. The value in AgenticOps isn't replacing human judgment, it is eliminating alert fatigue so engineers focus purely on high-leverage exceptions. Building explainability and controls directly into the infrastructure ensures agents act deterministically within defined operational bounds. Autonomy without verifiable guardrails is reckless, but speed without automation is impossible. A timely and essential framework.
I have a joke in my linkedIn about CISCO equiment dirt with pidgeon stuff( nest feathers and everything else) but the ligths remain green in the acces point and in the router, and I was thinking how this alert should look like in the CISCO management console, and how AI can work on it? How many alerts where an Agent can work analize and solve, to leave the real critical ones to a human. Now I know rhe answer 46%.
Great research. 4,100 alerts a day is an incredible amount for any team to digest! This is exactly where AI can make a real difference for our customers. Excited about what we’re doing with Cisco AgenticOps to help simplify that complexity while keeping organizations in control.