AI networking tools have many beneficial capabilities. When evaluating AI tools, engineers should prioritize the specific capabilities their networks need for successful AI use.
AI tools are prevalent in most industries, and all include a vast array of capabilities -- but network engineers might not want or even need all of them.
Every network has unique needs and supports specific business cases. Network engineers and developers must therefore ensure their AI tool sets have the capabilities their networks need. Network teams tend to want similar capabilities when it comes to AI, and they face common challenges as well when integrating it within their networks.
Rise of AI
AI tools aren't new. According to Jim Frey, principal analyst for networking at TechTarget's Enterprise Strategy Group, versions of AI have existed for decades via automated and expert systems. But AI is only now receiving more attention.
"AI has been around for a long time, but the interesting thing is, only a minority -- not even half -- [of people] have really said they're using it effectively in production for the last three years," Frey said.
Frey credited generative AI with AI's boost in popularity. Because GenAI is a newer technology with different capabilities, it's distinct from other AI tools in its use and deployment. Shamus McGillicuddy, vice president of research at Enterprise Management Associates, divided AI tools into two categories: GenAI and AIOps.
"There's generative AI, like ChatGPT, and that's recently become more popular, something that IT people in general are talking about," McGillicuddy said. "And then there's AIOps, which is machine learning and anomaly detection and analytics."
Complex and dynamic networks are also driving the rise of AI in networking, Frey said. He added that humans can no longer manually keep up with network demands and environments, but AI engines can.
What AI tool capabilities do networks need?
Though individual networks have different needs, many engineers want similar capabilities when they integrate AI into their networks. Experts agree that some of the most popular AI capabilities network engineers and managers want include the following:
Network performance optimization.
Predictive capabilities, including maintenance and monitoring.
According to McGillicuddy's research, network optimization and automated troubleshooting are the most popular use cases for AI. However, network professionals still want to fix the problem manually.
"Automated troubleshooting [can] isolate the problem, analyze what the cause of it is and then fix it," McGillicuddy said. "Usually with the fixing part, people don't want the machine to do it on its own. They want machines to present the fix, then they approve it."
I don't know if you can put one [capability] ahead of the other. It depends on what tools they're using and how effective those tools have been.
Amy Larsen DeCarloPrincipal analyst for security and data center services, GlobalData
Many of these capabilities can help mitigate security risks and threats, and according to Frey, many networking professionals look at AI as a way for organizations to get better and smarter about security. Amy Larsen DeCarlo, principal analyst for security and data center services at GlobalData, agreed.
"Network managers are looking for the same things that a security professional would be looking for in the sense of being able to recognize the problem and better address it before it becomes an outage," DeCarlo said.
Outside of security functions, Frey also listed some alternate use cases for AI, including documentation and change recommendations. Though less popular than other features, Frey explained that recommendations become some of the most valuable capabilities for network teams. But no matter the capability, every tool has a place in a network environment if it can fit the needs of the network and network team.
"I don't know if you can put one [capability] ahead of the other," DeCarlo said. "It depends on what tools they're using and how effective those tools have been."
Generative AI
Despite being a more recent development, GenAI has quickly made itself useful within the networking world. McGillicuddy said that in the past year and a half, network professionals have started using GenAI tools and found them useful. One of the most popular tools is also the most well-known: ChatGPT.
"One guy told me that doing stuff with ChatGPT could cut a task from four hours to 10 minutes in some cases," McGillicuddy said. However, he also warned that network professionals should be knowledgeable about the tools they're using, as GenAI tools can easily make mistakes.
"It's very possible it's going to do something wrong, hallucinate, and if you just have absolute faith in the type of content something like ChatGPT will create, you could mess up the network really badly," he said.
Aside from ChatGPT, vendors are also creating GenAI interfaces for their products, such as virtual assistants. McGillicuddy's research found that the most popular use cases for vendor GenAI products are the following:
Querying IT systems.
Querying product documentation.
Automating actions.
DeCarlo also said that an additional benefit of GenAI tools is their training capabilities. Because of the tools' faster speeds and ability to go deeper into the network, she said they can offer potentially valuable insight into the network, and therefore expedite the learning process.
Regardless of origin or use, Frey credited GenAI's overall popularity with its ability to outperform older systems because of those systems' lack of sophistication. However, GenAI's complicated infrastructure, high speeds and performance sensitivity have also helped give rise to AIOps tools, which are needed to support it.
"We're not going to be able to manage these GenAI infrastructures without the help of AI tools, because humans won't be able to keep up with the change," Frey said.
Challenges of AI tools
Though AI tools can be beneficial to networks, network engineers and managers should be aware of some challenges before implementing them.
Data privacy, collection and quality
Concern over data and its usage is one of the biggest issues for those considering integrating both AIOps and GenAI tools. Frey said that because of the mixed nature of network data -- which includes both anonymous, general operational data and personally identifiable information -- data privacy is one of the top concerns.
For GenAI specifically, McGillicuddy said that primary concerns revolve around quality validation for AI outputs, as well as using good quality data to train AI tools. "If you're feeding bad data to a generative AI solution by one of your vendors, it's going to have problems understanding what is going on in your network."
Complexity
Both Frey and McGillicuddy agreed that AI and network complexity could be a hinderance to AI tools. Frey explained that AI systems, especially GenAI, need tuning and good recommendations to avoid hallucinations and bad recommendations.
On the network side, McGillicuddy said complex network infrastructure could hinder some AI tools. "[Enterprises have] really complex, maybe multi-vendor, networks, [and] the AIOps components are only smart about certain pieces of it," he said.
User uptake and skills gaps
Another major challenge has less to do with AI technology and more with the network teams using it. User uptake isn't guaranteed, nor is a team's ability to use the AI products, so proper training is necessary to reap all the benefits of AI in networks.
Some network professionals don't want to use AI at all, while others don't understand how to integrate it into their networks. McGillicuddy said AIOps in particular is less user-friendly than GenAI, so having some AI skills is necessary to get value out of the technology.
"Understanding how tools work and where there might be gaps can be a challenge," DeCarlo said. She further explained that the learning curve for using the tools can be steep, especially if network teams have been using other tools for a long time.
Integration
Part of the issue with user uptake and skills gaps is integration. McGillicuddy said there are two sides to the integration problem: tools and processes. On the tools side, he cited concerns about integrating GenAI tools with tools already in use and getting them to work in harmony.
"On the process side, it's about making sure people use this in an effective way," McGillicuddy said.
Aside from vendor AI tools, DeCarlo warned that organizations might have to develop and integrate in-house supplemental tools. It can also be challenging to integrate vendor AI with tools teams have created and customized themselves.
Future of AI tools
Organizations are already integrating AI into their networks, and the future of AI tools in networking is as strong as ever.
When it comes to GenAI, Frey said he believes organizations and network engineers are just now starting to see its real value, though there's some uncertainty about where it will make a difference. However, he also stated that early use shows it's helping to improve predictive capabilities. These predictions could enable network engineers to be more proactive when it comes to addressing network problems.
"In reality, most [of] network operations is a reactive practice," Frey said. "You set things up, hope they're okay, but then something goes wrong, and you don't know about it until someone's been affected."
GenAI's ability to take instructions and roll out network changes could also be a promising feature, though McGillicuddy said vendors might be slower to adopt these types of interfaces. He explained that with thorough instructions, these GenAI interfaces can speed processes up and empower those who might not know enough to roll out network changes properly.
DeCarlo argued that AI tool capabilities should focus more on integration between security and network management going forward. "Security can impact network performance," she said. "If you can identify a potential security problem, you can avoid network outages."
These aren't the only directions in which AI might develop. Both GenAI and AIOps have limitless potential, some of which developers and network engineers haven't even scratched the surface of yet.
"There are probably use cases we haven't developed yet, and as networks evolve, we may need different capabilities than we do today," DeCarlo said.
Nicole Viera is assistant site editor for TechTarget's Networking site. She joined TechTarget as an editor and writer in 2024.
