“Agentic AI infrastructure for network operations is neither a task for CSP DIY nor for AI generalists, but rather for a relatively handful of players committed to a long-term roadmap centered on network domain expertise.” – Grant Lenahan, Principal Analyst, Autonomous Networks

This Appledore Research whitepaper, authored by Principal Analyst Grant Lenahan and produced in partnership with Nokia, examines what it actually takes to deploy effective agentic AI in IP network operations. Written against the backdrop of an industry moving rapidly from AI hype to engineering reality, it draws on more than a year of primary research — dozens of deep briefings with CSPs, network suppliers, ISVs, and hyperscalers — to identify a clear and emerging consensus on the right path forward.

We assess the four interdependent pillars on which successful network agentic AI must rest:

• intent-based control theory
• domain-specific structured data and ontologies
• digital twins
• agentic infrastructure including MCP-native APIs and an agentic studio.

We argue that focusing on which LLM to use, or turning generic AI loose on existing data lakes, is the wrong path — and that the correct path demands deep network domain expertise. We also provide our candid view on the supplier landscape, distinguishing domain experts, NAS OSS ISVs, hyperscalers, and DIY approaches across a set of clear capability dimensions.

This paper is essential reading for CSP network operations leaders, OSS/BSS architects, and CTO-office strategists charged with turning agentic AI investment into durable operational gains.