Just like your old maths teacher, Cisco’s test for autonomous network agents is to show their working

Résumé détaillé

Cisco executive Joe Vaccaro says AI agents need to demonstrate the evidence behind a proposed fix and operate within established procedures. For network engineers, that makes...

Cisco executive Joe Vaccaro says AI agents need to demonstrate the evidence behind a proposed fix and operate within established procedures. For network engineers, that makes documenting their own expertise increasingly valuable.

Before an AI agent is trusted to change a production network, Joe Vaccaro wants it to satisfy a requirement familiar to any teacher: show the work behind the answer.

Speaking with iTWire at Cisco Live in Melbourne, Cisco’s senior vice president and general manager of network platform and assurance describes trust as something that must be established throughout an agent’s work.

Operators need to understand which signals led to a diagnosis, what evidence supports the proposed cause, and why the selected workflow is appropriate. A plausible recommendation is helpful, but it's really only the beginning when the next step could affect a working network. You absolutely must have trust to let the agent go further.

However, to go further and to much more advanced, autonomous agentic network activity, the confidence and risk profile may well be very different. Here's where "showing the working" helps explain why an agent made a recommendation and the evidence and information it used along the way. This all combines to give trust.

It's essential too; Cisco research in conjunction with Omdia identified that 46% of network alerts are closed without investigation because of staffing constraints, recurring noise, or false positives. The research surveyed more than a thousand IT and network operations leaders.

Something has to change if organisations want to keep on top of their network tickets, and Cisco sees the solution as AgenticOps.

Vaccaro describes Cisco’s AgenticOps approach as a sequence that begins with sensing a problem, establishes where it resides, develops a remediation plan, validates the proposed change, and then deploys it.

Within that sequence, Cisco uses deterministic execution workflows informed by the customer’s existing policies and…

Début de l'article, extrait de iTWire. Pour lire la suite, ouvrez l'article chez le média.