Network staff skill sets lag when it comes to running AI and automation on networks. This comes at a time when companies expect instantaneous network issue resolution, as well as the ability to scale and deploy as many new applications internally and in the cloud as they can -- often in a matter of seconds. Developers already have highly automated deployment methodologies, but network professionals trail behind.
It’s time to look at the next wave of automation for network operations. In other words, can the deployment of network AI agents and automation be used to speed up network deployment cycles and issue resolution?
AI network agents are in exploratory stages of deployment. Their ultimate goal is to extend network automation beyond observability and AIOps, into a province where the fundamentals of network management are as fully automated as possible. This includes monitoring, alerting, responding and resolving incidents, as well as applying enterprise security and compliance. Once confidence is gained in these agents' capabilities, the next step would be for these agents to automatically scale and manage network resources so that application workloads can be optimized.
Related:Avoiding network logjams in the age of AI
To do these tasks, network AI agents require a set of business rules from the network, security and compliance teams. From here, the AI agents use machine learning to understand the network so they can self-improve their performance.
Today, a combination of factors has contributed to the deployment of AI network agents being more aspirational than actual.
While network staff can create the business rules for network management and scalability that AI network agents need, they must also ensure that these rules and guidelines are uniform across all networks, whether networks are in the data center, at the edge, or in the cloud. Many sites struggle with this because they have so many diverse networks.
There are also issues with system and network integration, and with the coordination of security, compliance and network management, which can span several different functional departments within the company. Collaboration across these teams can be challenging in practice -- but without it, AI agents are left with gaps in their instructions.
Just as agents must be trained, there are also staff learning curves when it comes to AI and automation. While most enterprise network teams have moved beyond standard network monitoring to observability, they are still only mildly engaged with AIOps, which is a critical stepping stone to network AI agents. The good news is that several major network vendors offer clear paths of technology migration that organizations can follow -- paths that can take sites from standard network monitoring all the way to network AI agents.
Related:Is your network infrastructure ready for AI workloads?
As network teams investigate AI network agents, they want to understand how these agents work and how network AI agents can deliver benefits to network operations.
In one AI network agent trial conducted in November, Nanites, which provides composable (i.e., modular) network AI agents, found the following:
"We simulated an interface outage across a Cisco IS-IS (intermediate system to intermediate system) network," said Nanites. "Nanites AI analyzed the alert and remediated in 3 minutes, a task that typically takes a skilled engineer 30+ minutes. Under the hood, the system did the following:
Autonomously handled an alert from Grafana (a metrics, logging and tracing software).
Identified the root causes through reasoning (likely by observing network patterns, configurations, topology and traffic, and then drawing a conclusion), not just rules or playbooks.
Determined precise troubleshooting steps dynamically in real-time.
Executed those steps autonomously, interfacing directly with systems.
Applied fixes in seconds (with human approval only)*
Related:It's time to revamp IT security to deal with AI
The asterisk reflects a footnote to the study, which states that this trial was conducted in a highly controlled network environment.
The time to resolution in the trial was impressive; at first glance, the AI agent was evidently much quicker than a human engineer would be. But the insistence on running the trial in a highly controlled network environment was equally noteworthy, as was the need for a network staff member to make the final decision.
Even in perfect systems – which is not the reality for most – the AI was not trusted to act entirely autonomously. This raises questions around how much human involvement is still needed if an organization hands its network management to AI agents.
Even where there is trepidation, it’s clear to network managers that the tools that their staffs are using won’t hold up to the avalanches of data that they now see on a daily basis.
In February 2026, Neraj Kumar, director of solutions engineering for Solarwinds, referenced IDC research that revealed that 59% of organizations were investing in AIOps as a means of automating more network monitoring, but that 75% of organizations were still tied up with "keeping the lights on." Tool sprawl was one reason sites were having difficulty moving forward—but so were data overloads from incoming network data and telemetry.
"No CIO walks in on Monday and says, 'My environment is simpler than it was last year,'" Kumar said. “Hybrid and multi-cloud adoption has given teams more flexibility, but also more integration points, telemetry streams, and ways for incidents to ripple across the stack."
Clearly, more AI and automation are needed to keep networks running and to scale them to tasks. This encourages the adoption of network AI agents to handle more work -- but are network staffs ready to deploy them effectively?
There is good news for organizations that currently don’t feel prepared to embrace agentic AI: This is actually a great time to lay the groundwork for AI network agents because the technology is still in very early stages of adoption. As long as those steps are taken now, network management shouldn't fall behind.
Here are four recommendations:
The network staff already knows that it’s getting inundated with overwhelming amounts of data and alerts as the network continues to scale. It also knows that it can't keep pace with every alert and that the tools in place can't always do the job. As a result, almost everyone will agree that more automation for network operations is needed, whether it comes from AIOps or network AI agents.
This is where the network staff should begin its evaluation. If you could automate any operations on the network, what are the operations that you would most want to automate with AI? What performance improvements would you expect? By setting clear priorities, staff narrow their focus and organize the strategy around meaningful business outcomes.
2. Define a strategy and construct a roadmap.
Once the network staff has defined its network automation and performance goals, the next step is to create a timeline for these improvements and to identify the technologies that can deliver the desired outcomes.
It would be great to imagine that a fully autonomous network using network AI agents could single-handedly run the network and deliver all of the performance goals, but almost no one would say yes to this. The Nanites AI agent trial is a perfect example: Performance was delivered, but only in a highly controlled network environment, with a human network professional standing by to make the final decision on which AI network agents to recommend.
Teams should keep this in mind when laying out their strategy. Network staff should consider how everyday friction in the system may affect AI efficiency and design roadmaps that account for the need for a human in the loop.
3. Partner with a forward-thinking network vendor.
Universally, enterprises and vendors see network management evolving from standard monitoring to observability to AIOps to AI network agents. However, not every vendor is equal when it comes to being a good business partner and having an effective technology roadmap for its products. Sites should take longevity into account when determining which network vendors to partner with; the aim is to find vendors that continually invest in their products, stand by them, and offer great support.
4. Trial AI network agents in controlled network environments.
Nanites trialed AI network agents in a highly controlled network environment. This enabled it to tailor its use case to observe how a set of network AI agents performed in a specific context. In other words, the trial wasn't conducted in the hybrid constellation of multiple cloud and internal networks that most enterprises have. Sites should learn from this. Go slowly at first by trialing network AI and automation in a highly controlled network environment. Once those kinks are ironed out, agentic AI can be tested in new domains and scale from there.
President of Transworld Data
Mary E. Shacklett is an internationally recognized technology commentator and President of Transworld Data, a marketing and technology services firm. Prior to founding her own company, she was Vice President of Product Research and Software Development for Summit Information Systems, a computer software company; and Vice President of Strategic Planning and Technology at FSI International, a multinational manufacturer in the semiconductor industry.
Mary has business experience in Europe, Japan, and the Pacific Rim. She has a BS degree from the University of Wisconsin and an MA from the University of Southern California, where she taught for several years. She is listed in Who's Who Worldwide and in Who's Who in the Computer Industry.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。