Agentic AI Is Already Acting on Production Networks, Cisco Survey Finds

Network operations teams are beginning to move beyond AI that identifies problems toward agents that can investigate and act on them. In an Omdia-conducted survey of 1,000 IT and network operations leaders, Cisco reports substantial production adoption—but also a clear condition: operators need to understand and control what agents do.23

How far has agentic AI reached production?

Cisco says 75% of surveyed organizations have deployed AI for network operations, and 51% already run agentic AI systems that take action in production rather than only offer recommendations. Another 84% expect to reach an AI-led operating model within 12 months. That last figure is an expectation, not a measure of deployments completed.27

Cisco describes the distinction from conventional AIOps as a move beyond alerts and recommendations: agents can reason through a problem and take action, with oversight from human operators.5 Production use, however, does not mean every organization gives agents unrestricted authority.

Why existing AIOps is falling short

According to Cisco, 95% of surveyed enterprises say their current AIOps tools cannot keep up.3 The reported average is about 4,100 monitoring alerts per organization each day, more than half of them network-related. Cisco estimates that manually clearing the network-alert portion of that volume would require roughly 100 IT specialists, based on an estimate of about 21 alerts handled per practitioner per day.228

The challenge is not just the number of alerts. Troubleshooting can require evidence from networks, cloud services, internet connections and security systems. An agent that correlates those signals may help an operator investigate a problem and decide what to do next; another isolated source of alerts would not solve the visibility problem.57

What changes in day-to-day operations?

The proposed workflow shifts some work from manually sorting alerts and moving between tools to supervising an investigation: an agent detects an anomaly, correlates telemetry, recommends or carries out a response, and checks the outcome. Cisco describes these as capabilities of its AgenticOps approach, not as proven time savings across every organization in the survey.78

AI workloads could also increase the pressure on networks. In a separate projection, Cisco says enterprise traffic could grow ninefold by 2035 with agentic AI adoption. That is a forecast about possible traffic growth, not traffic measured by the Omdia network-operations survey.10

Autonomy still needs an audit trail and human control

Cisco reports that four in five respondents are comfortable giving agents a highly or fully autonomous role, including nearly a quarter who would allow action without a human in the loop. At the same time, more than two-thirds require detailed explanations of agent-driven actions, and 86% see a single integrated platform as more effective than another point tool.34

In practical terms, operators need to see what an agent observed, why it proposed a change, whether approval was required, what it did and whether the action worked. Policy limits, approval points and an emergency override help make that authority governable; independent coverage of the survey reports that respondents overwhelmingly expect such guardrails.817

The takeaway: the survey shows meaningful reported adoption of agents that act in production, driven by an alert burden existing tools struggle to manage. It does not establish that every deployment has reduced incident duration or staffing needs. The next test is whether teams can verify agents’ decisions and retain appropriate control as those agents take on more operational work.238