The progression of AIOps in today’s network environment

The path to network autonomy

At Nokia, we’ve helped design and build many of the largest networks on the planet. Alongside that, we’ve built operations tools that help optimize the design, deployment, and operation of those networks. Network management tools have always played a critical role in unlocking the most capability and value from networks of all types, around the world.

In recent years, the growing demands of scale and complexity have driven interest in Autonomous Networking, where the tools used to manage network infrastructure take on a more sophisticated role, promises to offload much of the mundane while increasing accuracy. The goal, of course, is for systems to perform the work of network management on behalf of the people who define the outcomes. Several autonomous network frameworks exist, with perhaps the most well know from the TM Forum.

Much like the levels of a self-driving vehicle, the TM Forum Autonomous Networks framework defines a number of levels, correlating to the amount of work performed by humans versus systems. From a completely manual starting point, each level indicates an increasing amount of overall work in the domain of a system – first with humans in the loop, and ultimately a completely system-driven management.

Most of the original thinking about autonomous networks didn’t consider artificial intelligence – at least not in today’s form. Certainly machine learning has long been a core feature of network management tools, but today’s large language model (LLM) capabilities add a significant new enabler on the path to network autonomy. “AIOps” appears frequently now, making a promise that AI somehow changes our fundamental capabilities in operating networks.

What enables AIOps?

The first, and perhaps biggest impact of AI is the ability to interact with our networks with natural language. Today’s network practitioners are expected to be linguists – deploying knowledge of a variety of interface languages from command line to database querying. The arrival of a universal natural language interface for network operations has changed the way we think about tooling. Inputs and outputs can now be more fluent, precise, and customized. Workforce impacts are also clear, allowing a greater pool of applicants by lowering the burden of knowledge to get started and become productive.

Secondly, the advent of agentic tool calling allows AI-enabled systems to interact with other systems in a programmatic fashion. We can quickly process logs, check configurations, and interact with ticket systems simply by enabling tools via MCP or API. Each of these tools is specialized, ensuring the accuracy of information as it’s processed. Much like assembling a great sports team, each position player has a job to do, brought together by the coach calling the plays.

The third enabler is the AI models themselves. True autonomy in network operations means that the systems operating on our behalf are at least as good as a human. Output from a model should be accurate, provable, and explainable. A model that excels at natural language understanding may not give the most accurate results for diagnosis, or configuration. Using multiple models in a workflow is common today, this can mean using small trained models for high quality, reproducible results – and leveraging large language models to interact with humans and their intent.

AIOps in Nokia's EDA for data center networks 

Nokia’s EDA is a Kubernetes-based, model-driven platform for managing large scale, complex data center networks. From deployment to ongoing maintenance, EDA simplifies management of both traditional data center workloads as well as the most advanced AI fabrics. As we looked to bring AI capabilities to EDA, we focused on high-impact use cases that leveraged the best of AI’s capabilities today.

We first took the lowest hanging fruit – the EDA Query Language (EQL). This powerful capability allows users to quickly find and filter information about the network with a simple query box. And while EQL has tremendous power, it’s still another language to learn. Our team leveraged AI to allow EDA users to form their question as natural language – “show me all down interfaces which are not disabled”, reducing the clock cycle of troubleshooting greatly.

In an upcoming release of EDA, we’ve added AskEDA – a chatbot companion inside the EDA user interface. AskEDA is a powerful tool which allows you to ask questions about what you’re seeing on the screen, quickly sort through complex information like alarms, and more. See something unusual you want to monitor on a regular basis? Search, sort, and filter information using AskEDA, and then have a customized dashboard widget created automatically. Need to triage the root cause of an issue? AskEDA will sort through logs and alarms and give you an evidence-based explanation of what’s wrong, and how to fix it.

We recently showed this capability at Network Field Day 39 in November of 2025. A video is worth 1,000,000 words – go check it out!

The path forward

AI’s impact on network operations is real, and at each level of the Autonomous Networks framework we will see it’s impact. Today, AIOps is a tool that can be leveraged as needed and appropriate to assist humans in the operation of large, complex networks. For an advance look at some of how we explore the design space with AIOps, check out this recent Packet Pushers VideoByte for more. We are excited to bring these new capabilities to market soon, and look for expanded capabilities powered by AIOps during 2026!