Nokia adds AI agents to network management software, promising operators they will keep control
Nokia has updated its Network Services Platform to include an agentic AI framework that can diagnose faults and suggest fixes in complex IP networks. The system, due for commercial release by the end of 2026, is designed to let AI act autonomously only within boundaries set by the operator, in response to persistent industry unease about handing control to automated systems.
Text by Martti Asikainen, 17.6.2026 | Photo by Nokia
Nokia has introduced an agentic AI framework into its Network Services Platform (NSP), the software used by telecommunications operators to manage large, multi-vendor IP networks, marking what the Finnish company described as a step towards fully autonomous network operations.
The announcement, made on 11 June from Espoo, focuses on the first practical application of the framework: an AI agent designed to identify the root causes of network faults more quickly than human engineers working alone.
Agentic AI refers to systems capable of taking sequences of actions and making decisions without requiring human input at each step. Applied to network management, this means software that can observe a fault, gather relevant data from across a network, reason about likely causes, and propose or execute a fix, rather than simply flagging an alert for an engineer to investigate manually.
What the system does
The NSP framework grounds AI agents in a continuously updated model of the network, covering topology, configuration state, protocol behaviour, and recent changes.
According to Nokia, this means agents reason from accurate, live data rather than from incomplete or outdated information. The company argues that data quality is essential before any automated system can be trusted to act in a live network.
The first agent built on the framework handles troubleshooting. Rather than presenting engineers with a raw stream of alerts, it analyses incoming fault data, identifies probable root causes, and presents its reasoning in a form engineers can follow and verify. Nokia says the approach reduces what the industry calls “operational noise”: the volume of alerts generated by cascading failures, many of which point to the same underlying problem.
The framework also supports communication with AI agents from other vendors via the Model Context Protocol (MCP), an emerging open standard for agent-to-agent communication. Nokia says this is intended to allow its agents to operate across the mixed, multi-vendor environments that most large network operators run.
The trust question
Nokia’s announcement is explicitly framed around operator reluctance to cede control to automated systems. The company says agents can only act within policies and access controls defined by the operator, and that all actions are intended to be explainable, meaning the system can show its reasoning, not merely its conclusions.
Sasa Nijemcevic, Vice President and General Manager for IP Network Automation Software at Nokia, said trust remained the central obstacle to wider AI adoption in network operations. The company’s approach, he said, was designed to respect how networks are actually operated, starting with high-impact use cases such as troubleshooting and expanding the role of AI incrementally as operator confidence grows.
Grant Lenahan, Partner and Principal Analyst at Appledore Research, an independent telecoms advisory firm, said the emphasis on data quality was well-placed. “Domain expertise is likely the most critical quality in designing effective automation for complex networks,” Lenahan said, adding that accurate data and clearly defined relationships between network elements matter more than the choice of AI model.
Availability and next steps
Nokia argues that IP networks are growing in size and complexity as AI-related data traffic increases, putting pressure on operators to automate tasks that previously required skilled engineers. At the same time, network faults increasingly carry commercial consequences measured in minutes: an outage affecting a cloud provider’s customers, for example, can ripple across thousands of dependent services.
Nokia positions NSP as the existing authoritative controller for IP networks, meaning the platform already holds accurate network state information and enforces policy. Adding AI agents directly to that platform, rather than layering them on top as a separate system, is the company’s argument for why its approach is safer than alternatives that would require AI to learn the network from scratch.
The agentic framework will be commercially available by the end of 2026, according to Nokia. The company says the troubleshooting agent is the first of what it expects to be multiple use cases built on the same framework, with operators able to add AI capabilities incrementally as confidence grows.
Nokia describes the intended end state as “fully autonomous networks”: systems capable of managing themselves without routine human intervention. The company did not set a timeline for that goal, and the gap between a guided troubleshooting agent and a fully self-managing network remains considerable.
The announcement comes entirely from Nokia’s own newsroom and does not include responses from network operators who have evaluated or deployed the system. The troubleshooting agent is described as the first use case to ship; whether operators adopt it at scale, and how quickly trust in autonomous action develops in live production environments, will determine whether the broader autonomous network ambition advances on the schedule Nokia implies.
