Generative AI is no longer just a developer's playground; it is rapidly becoming a "monstrous data processing tool" for the enterprise, according to one industry analyst.
However, as organizations move from pilot projects to production, the transition from deterministic code to probabilistic AI models is creating a new set of challenges for IT operations and observability teams.
In this episode of the IT Ops Query podcast, Beth Pariseau, senior news writer at Informa TechTarget, spoke with Torsten Volk, principal analyst for application modernization at Omdia, a division of Informa TechTarget. Volk shared his insights on how generative AI (GenAI) is reshaping application modernization, the rise of agentic AI, and why the "human-in-the-loop" remains the most critical component of a successful AI strategy.
Business transformation with AI agents
Torsten Volk
Volk identified two primary ways GenAI is influencing application modernization. First, cloud service providers and consulting firms are deploying agent frameworks to automate everything from discovery to testing. This is particularly valuable for legacy systems, such as mainframe COBOL code, where AI agents can assist in documenting and decomposing monolithic applications into microservices.
"If you use AI agents to help you with the discovery, then that is tremendously useful," Volk said. "The key is that you understand what the limits are, and the key differentiator really is to make those AI agents talk to the human."
The second shift involves developers incorporating GenAI capabilities directly into software. This enables higher levels of customization and probabilistic functionality that weren't possible with traditional rule-based programming. However, this shift requires operations teams to rethink how they monitor and secure applications that may not produce the same result twice.
The observability bottleneck
The AI knows my intent ... and can pre-qualify what an observability system serves up.
Torsten VolkPrincipal Analyst, Omdia
For observability teams, the promise of GenAI lies in its ability to contextualize the "flood of alerts" that often overwhelms human operators, Volk said. By processing massive data streams in near real time, AI can pre-qualify incidents based on business intent -- such as maintaining revenue or customer service goals -- rather than just technical metrics.
"The fact that a button or an alert is red versus orange versus green doesn't really mean anything, because maybe I don't care about something being down on a specific day," Volk explained. "The AI knows my intent ... and can pre-qualify what an observability system serves up."
Yet, monitoring the AI itself presents a new frontier. Traditional metrics like latency remain important, but AI observability must now account for hallucinations, data degradation, and even legal or financial risks when an agent provides incorrect advice.
The future of agents
While there is significant hype surrounding agentic AI -- where multiple AI agents work together to solve complex tasks -- Volk warned that enterprise adoption is still in its early stages. The challenge lies in the probabilistic uncertainty that compounds as agentic AI workflows grow longer.
"The longer those workflows get, the less you can predict the results, and the more risky all of that gets," Volk said. "We are scratching the tip of the iceberg."
Emerging standards such as the Model Context Protocol and Google's Agent2Agent may eventually provide the orchestration needed for enterprise-scale deployments, much as OpenTelemetry did for cloud-native monitoring. Until then, Volk advises organizations to focus on transparency and to maintain a subject matter expert at every stage of the process.
Editor's note: An editor used AI tools to aid in the generation of this article. Our expert editors always review and edit content before publishing.
Torsten Volk
