Cisco (NASDAQ: CSCO) says enterprise network teams are moving agentic AI from advisory tasks into production operations, but its latest research also identifies visibility and explainability as the controls organizations want before granting those systems broader authority. The central issue for CIOs is not simply whether an AI agent can take action, but whether operators can understand, audit and reverse that action when it affects a live network.
Cisco’s report, The Impact of Agentic AI on Network Operations, is based on an Omdia survey of 1,000 IT and network-operations decision-makers at organizations with at least 500 employees in North America, Western Europe and Asia-Pacific. The company said 75% of respondents have deployed AI for network operations, while 51% run agentic AI that acts in production today. Those figures reflect respondents’ reported deployments, not an independent audit of their environments.
The report frames the adoption pressure in operational terms. Cisco said the average organization generates about 4,100 monitoring alerts and events each day, more than half of them network-related. It estimates that clearing a daily network-alert backlog manually would require roughly 100 IT specialists. The company also said respondents reported that nearly half of network alerts are closed without investigation and a similar share of investigation time goes to false positives.
That volume helps explain the appeal of AI that can correlate signals, make recommendations and execute bounded changes. Cisco said respondents cited use cases including rerouting traffic, adjusting wireless parameters, isolating suspicious endpoints and resolving incidents without prior human approval. But an agent that can alter production systems shifts the operating model from using AI as a dashboard assistant to governing it as a participant in change management.
The survey found that 80% of respondents are comfortable giving AI a high or fully autonomous role in network operations, including 24% comfortable with no human oversight. At the same time, 69% require detailed explainability for agent-driven actions, and 36% said full observability, including traces, rationale and post-action audits, is the minimum acceptable standard. Cisco said 86% favor a single integrated platform over additional point tools.
Those expectations signal a practical constraint for enterprise adoption. An automated action may restore service faster, but the operation still needs an evidence trail: what data informed the action, which policy allowed it, what systems changed and whether the outcome was validated. Without that record, operations teams can trade an alert backlog for a governance and troubleshooting problem that is harder to manage.
AI workloads add another reason to improve that operational foundation. Cisco said its aggregated direct-to-AI network telemetry indicates AI traffic is on a trajectory to double every six months. In Cisco testing, it said tasks performed by agents can generate up to 450% more total network traffic when agentic AI traffic is included. The company did not provide methodology details for those telemetry and testing figures in its published summary.
For network leaders, the immediate takeaway is to define which decisions an agent can make, which require review and which must stay under human approval before increasing autonomy. That means connecting observability, change controls and escalation paths rather than treating an AI assistant as a standalone tool. Legacy NetOps models often split monitoring, ticketing and remediation across separate systems; Cisco’s findings suggest that the competitive advantage from agentic operations will depend on integrating those controls well enough to make automation accountable as well as fast.