Nabugu - stock.adobe.com
Why AI in network operations requires human judgment
Many network pros are skeptical about AI, but it's most effective as an assistant to human judgment. A framework helps engineers improve skills and build confidence.
Many network professionals are skeptical about AI. However, AI is less of an antagonist and more of an assistant to network professionals, and it works far better when users work alongside AI rather than against it.
During this week's Spring 2026 AI Networking Summit, presented by ONUG, IBM outbound product management leader Jason Lovelace presented on the importance of this partnership. AI is being used across every sector of work, but how network professionals engage with the technology differs from how other professionals use it, Lovelace said. Even if AI is accurate 80% of the time, that's not sufficient for networking.
"To implement AI in the network, you need to think about maintaining or improving on three nines," Lovelace said. "The goal here is the illumination of human judgment through AI as a partner."
To achieve this high availability, organizations shouldn’t give AI free rein in the network. Rather, Lovelace suggested network professionals maintain their judgment and work with AI as a partner. He outlined this process through IBM's four-step framework:
- See: Gain temporal insights.
- Use: Apply reasonings and use tools.
- Prove: Evaluate suggestions and apply guardrails.
- Act: Execute the changes.
This article outlines how network professionals can follow IBM's model to improve their AI use in networking.
1. See
This first step of the model concerns how AI observes the network and gathers information from the environment. According to Lovelace, an agentic AI system needs two critical pieces to collect useful information about the network:
- Telemetry.
- Context.
Telemetry
Real-time telemetry data is essential for the AI tool to understand what happens in the network as soon as it occurs. Delays of up to a few minutes can become problematic because the system will no longer evaluate the network in real time.
Lovelace suggested using time-series monitoring -- an approach that collects metrics over time and uses past behavior to understand how a system will perform -- to evaluate telemetry data. Time-series tools can also spot issues within the network that occur before it passes the threshold before an alarm.
According to Lovelace, combining agentic AI with time-series monitoring enables the AI to have a comprehensive understanding of the network, the next most essential part of analyzing data.
Context
When an agentic AI model has deep context about the network, teams can use it to analyze anomalies and other issues that occur in the network more thoroughly, Lovelace said.
For example, network teams can use this metric data to determine whether an issue is isolated or points to a broader issue, such as distinguishing whether an alert is the issue itself or a symptom of the true problem.
"We need both of these things working together to give engineers a clear view of what's happening," he said.
However, Lovelace cautioned against an AI agent having too much context, as it could perform less accurately and lead to more problems than fixes.
2. Use
Network professionals next apply AI tools to the network. According to Lovelace, one of the most important things to recognize about this step is that the models are less important than the tools themselves.
As models continue to evolve, the type of model used is less important than the ability to create and secure useful tools for network engineers, he said. AI tools that generate code can especially help teams build as many tools as needed, giving network engineers a vast array of tools to support operations, he added.
"The richness of the tools is critical for thinking about what your network engineers do on a daily basis," Lovelace said.
3. Prove
This step is less about proving that AI works and more about demonstrating AI reliability. Unlike other issues that AI tools might help remediate, network problems are more likely to require more prompts in an AI tool. The more requests submitted to an AI tool, the more complexity it has and the lower its accuracy.
"If your context window is filling up, the compounding effect of a hallucination at step four or five then makes steps 17, 18 and 19 not reliable," Lovelace said.
Because of the complexity of network operations, engineers must prevent AI tools from relaying inaccurate information down the chain of processes. Additionally, network teams must implement guardrails to prevent compounding errors and ensure the network's overall reliability.
4. Act
The final step of the model defines how network engineers put AI into practice. Rather than offering a blanket approach to AI network operations, Lovelace said the ways network engineers use AI depend on their level of seniority.
Junior network engineers would benefit from using AI to guide their decision-making. AI can suggest steps, and network engineers would follow and implement them while learning the explanations. A senior network engineer, on the other hand, should consult AI for recommendations and then decide which, if any, of its suggestions to implement.
"They want to be able to have the ability to push back on the model and have the model reconsider its actions," Lovelace said.
AI framework builds trust across engineering teams
An AI framework could provide a means for AI use that network engineers and leaders are comfortable with. Additionally, it could help engineers improve their skills both directly within network operations and validate their judgment.
For junior network engineers, an AI framework could help them understand the work they're doing in a network when analyzing, validating and using network data. This understanding could help a network engineer advance in network operations, from entry-level to mid-level to senior-level. For senior network engineers, an AI framework can help them build trust in AI models and enhance their judgment and understanding of network behavior.
Finally, an AI usage framework can also ensure that leaders know their engineers are using AI safely, in line with governance and reliability requirements.
"If the AI technologies change, the framework allows organizations to reap the benefits of agentic NetOps and consume those changes while the standards of good governance and clean execution don't change," Lovelace said.
Deanna Darah is site editor for Informa TechTarget's SearchNetworking site.
