GenAI tools set to exceed past waves of AI automation

IT pros with experience in AI automation might hear a familiar echo amid the latest hype about generative AI and AI agents that can resolve or prevent IT incidents, but one analyst believes that newer tools are poised to surpass their AIOps predecessors.

AIOps -- or AI for IT operations -- tools that use machine learning to predict and respond to IT issues first entered the enterprise IT mainstream in 2018 and prompted some proponents to predict a future of hands-off, self-healing systems, or NoOps. But that future never materialized. Now, some of the same observability tool vendors are touting more intelligent AI automation tools using large language models and AI agents -- but has the industry seen this movie before?

Gregg Siegfried

Gartner analyst Gregg Siegfried said in a recent interview during Informa TechTarget's IT Ops Query podcast that he thinks not.

"Hopefully, agents will have more of a [clear] definition than AIOps ever did," Siegfried said. "Hopefully, the agents will be packaged in a way that they're more easily consumable, more easily understandable [and] more easily debuggable."

A broader trend in the industry of IT automation tool consolidation also portends easier management for AI agents than was available with AIOps tools, which often took a "manager of managers" approach to unify multiple disparate systems, Siegfried said.

"As a point of extensibility, or a point of augmentation, they're bringing in some agents from trusted vendors to serve a specific, limited or constrained purpose," he said of enterprise IT buyers. "I think [that] is a much more understandable value proposition for IT leaders to consume than trying to bring in some much larger platform."

They're bringing in some agents from trusted vendors to serve a specific, limited or constrained purpose. I think [that] is a much more understandable value proposition for IT leaders to consume than trying to bring in some much larger platform.

Another observability trend that could boost the effectiveness of AI automation tools is the separation and centralization of data management from tools that analyze and take action, such as the adoption of the OpenTelemetry standard, Siegfried said. This could help to avoid some of the data quality issues faced by AIOps early adopters.

However, generative AI also consumes data in disparate and unstructured formats more easily than past machine learning systems, he said.

"If you had an event correlation system that was taking events from five or 10 different monitoring tools and trying to correlate them, each one may have called the same thing by a different name," he said. "A lot of the heavy lifting you had to do was to try and map all those fields ... That's something that the generative AI engines are kind of doing naturally."

Still, organizational confusion and inertia about who is responsible for AI model performance are among the obstacles to enterprise adoption of AI automation tools, according to Siegfried.

"Is it generating hallucinations? Is it being attacked from a security perspective? What's the quality of the model? Those have normally not been the province of an I&O [infrastructure and operations] team or IT team," he said. "It's more the province of the data team that was responsible for the [model] training, but normally, in a lot of organizations, they're not going to be on call ... Understanding what job roles are going to be associated with generative AI applications is something that we've yet to quite figure out."

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.