Cisco highlights a small network team’s rule for AI: checking must cost less than doing

A three-person network team supporting 34 Room & Board locations delegates routine monitoring, wireless radio management and logging to agentic platforms, according to Cisco-sponsored coverage published November 11 on Harvard Business Review’s website. Senior network engineer Mark Rodrigue sets a condition: checking the agents’ work must cost less than doing it. The article revisits Cisco’s September survey.

From recommendations to network changes

Cisco describes agent-powered network operations as a model where people set direction and guardrails while software assesses conditions and takes action across systems. Those actions can include rerouting traffic, adjusting wireless settings or isolating a suspicious device—not merely recommending a fix.

The November article argues for setting permissions in advance: which actions agents may take alone, which require approval and which are prohibited. Requiring a person to approve everything, it warns, would turn an alert backlog into an approval backlog. Close to a third of surveyed organizations make multiple production-network changes daily.

That requires seeing beyond one system. In the survey, 92% said performance problems commonly span multiple domains and require correlating information across ten or more tools. The sponsored article cautions that an agent with a partial view might fix a visible symptom while making the underlying problem worse.

We don’t really have the time and resources to spend managing management tools. If your tools require a full-time babysitter, you’ve already lost.

Mark Rodrigue, senior network engineer at Room & Board, in Cisco-sponsored coverage on Harvard Business Review’s website

Adoption is not blanket permission

Cisco’s September 23 release said Omdia surveyed 1,000 IT and network operations leaders at organizations with at least 500 employees. Respondents came from North America, Western Europe and Asia-Pacific. Some 51% said their organizations ran agentic AI in production—self-reported adoption, rather than an independently verified census of enterprise networks.

The operational pressure is substantial. Respondents reported about 4,100 monitoring alerts and events daily per organization, more than half network-related. Cisco estimated that manually clearing the daily network-alert backlog would require roughly 100 IT specialists. The November article says nearly half of network alerts go unexamined.

Still, willingness to delegate selected tasks differs from removing oversight. While 82% were comfortable allowing some production-network changes without prior approval, only 24% were comfortable with fully autonomous operation without human oversight.

Respondents also specified what trust would require: 69% wanted detailed explanations for agent-driven actions. For 36%, the minimum included detailed activity records, summarized reasoning and post-action audits. Those are requirements respondents set, not evidence that deployed systems already meet them.