Service providers need a North Star architecture, embedded governance and agentic platforms that connect AI to real network context
For service providers, the agentic AI conversation tends to split into two tracks. The first is operational: use AI agents to automate workflows, reduce opex and move toward autonomous networking. The second is commercial: use AI-era infrastructure demand to create new service revenue. Cisco’s Rana El Desouky Kazamel argues those two tracks are not separate. They are mutually reinforcing.
Speaking at DTW Ignite in Copenhagen, Kazamel said service providers are “uniquely positioned” to monetize AI because they already have assets that matter in the inference era, including power, space, distributed infrastructure, customer relationships and proximity to users. As AI shifts from centralized training toward more distributed inference, operators can play a larger role in edge services and sovereign AI.
But that revenue opportunity depends on the operating foundation underneath it. “You need to build the right foundation with agentic ops efficiency, scaling the networks, operating them in an autonomous manner so that you can deliver services a lot faster, go to market faster,” Kazamel said.
That is the point where Cisco’s argument becomes more specific. Agentic AI is not a set of isolated use cases. It is a platform architecture. Kazamel’s advice to operators starts with defining a “North Star architecture” — the ideal end state — then moving quickly into use cases that can show ROI. The discipline is to avoid building disconnected pilots while still moving fast enough to prove value.
The data challenge follows the same logic. Operators do not need to wait until every data source is perfect before they begin. Kazamel pointed to data retrieval agents as a foundational element of Cisco’s agentic platform. “We’re going to meet you where the data is,” she said, so operators can connect to existing systems, complete use cases and show value faster.
That maps closely to Cisco’s broader Crosswork AI positioning. In a recent Cisco blog, Kazamel and Andy Schutz describe Cisco Crosswork AI as a secure, scalable multi-agentic framework integrated into Cisco Crosswork Network Automation. The framework is designed to act as an extension of the network team, with specialized agents reasoning through problems, identifying risks, troubleshooting issues, validating intent and recommending corrective action in real time.
The blog also highlights the platform elements needed for agentic operations to scale: agent evaluation, a knowledge graph, an agent catalog and an extensible model that supports both built-in and customized agents. Cisco positions Crosswork AI as part of its broader Agile Services Networking architecture and says it will also integrate with Cisco AI Canvas inside Cisco Cloud Control for unified, cross-domain management.
Kazamel further described change management as “the bigger hurdle,” arguing that the technology exists today but has to be embedded into the right processes, data sources, systems and ways of working.
Trust is a material blocker. Engineers will not use agents they do not understand, and operations leaders will not delegate work to systems they cannot govern. Kazamel said agents should be treated as extensions of the network operations team. “Just like humans, agents have an identity,” she said. “They have a set of guardrails…a set of policies,” and they have to be evaluated based on performance.
Agentic operations goes beyond automating tasks at machine speed. It is about making agents observable, secure, contextual and accountable enough to work inside high-consequence telecom environments. Cisco’s broader AgenticOps strategy similarly emphasizes intelligent execution with oversight, reliability, accuracy and governance at scale.
The agentic network, in this formulation, starts with architecture. Operators need a destination, but they also need near-term use cases. They need better data, but also mechanisms to retrieve and use the data they already have. They need AI agents, but also identity, policy, observability and evaluation. And they need efficiency gains, not as an end in themselves, but as the operating foundation for faster service creation.
Kazamel provided pragmatic insight: autonomous networks will not emerge from disconnected agents or isolated pilots. They will come from platforms that let operators standardize how agents are built, governed, connected to context and put to work.
