Many companies are still struggling to turn artificial intelligence pilots into real results, largely because scaling AI and moving toward agentic AI operations depends on solid foundations. At the same time, AI is following a familiar path seen in past tech waves, with early experimentation happening now, and broader impact as the right use cases emerge.

However, most enterprises lack the clean data backbone that makes AI deployment successful. That’s why AI prototypes might look good on the surface, relying on curated data that doesn’t match actual production conditions, according to Raju Penmetsa (pictured, right), chief technology officer of Fabrix.ai Inc.

Fabrix.ai’s Raju Penmetsa and Shailesh Manjrekar talk with theCUBE about the reasons companies fail to scale AI.

“That’s why you are not getting the results,” Penmetsa said. “I see that as [the] biggest problem, or biggest obstacle, our customers are seeing today.”

Penmetsa and Shailesh Manjrekar (left), chief AI and marketing officer of Fabrix.ai, spoke with theCUBE’s Dave Vellante and Bob Laliberte, as well as Zeus Kerravala, principal analyst at ZK Research, at the “Agentic AI Unleashed: The Future of Digital & IT Operations” event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed the gap between AI prototypes and real enterprise results and how agentic AI operations can overcome data fragmentation. (* Disclosure below.)

The foundations of agentic AI operations

Organizations need strong technology foundations to enable digital co-workers and agentic operations. This requires multiple fabrics, according to Penmetsa.

“One is our data fabric — which is the plumbing fabric for all of [the] different data sets you have — dynamically bringing together data of different types with the enrichments and that dynamism in that nature for a system to query that data,” he said.

The AI-enhanced fabric applies reasoning on top of the data to determine the appropriate action. Then, the automation fabric executes that action across whatever environment is needed, bringing all the components together into a unified system, Penmetsa explained.

“The most important thing is [that] AI is a co-enabler for you with that reasoning … versus giving you a decision or giving you insight for you to act upon it,” Penmetsa said. “That’s how we are seeing the shift from AIOps to the agentic operations.”

Of course, the key difference between AIOps and agentic ops is the “last-mile problem,” according to Manjrekar. With AIOps, humans still interpret tickets and dashboards, but agentic ops puts a large language model or agent at the center of decision-making, allowing it to reason and take action autonomously.

“What that really means is you’ve got to have the right building blocks,” Manjrekar said. “LLMs are just a means to an end.”

Fabrix.ai operationalizes this through its agentic ops framework, which is a core part of its value proposition, according to Manjrekar. It takes a prompt or intent, applies the right AI personas and Model Context Protocol tools and then enforces guardrails to keep LLMs from hallucinating.

“As much as there is promise of LLMs, you’ve got to have the right guardrails around it to be able to implement this at a scalable enterprise scale level, and that’s what Fabrix does,” Manjrekar said.

To achieve these guardrails, Fabrix.ai’s use of MCP taps into different types of data without having to move it around, according to Penmetsa. The system then identifies only the key attributes it needs and retrieves them iteratively, using a smart cache to maintain the right amount of context.

“That’s how we are solving it in Fabrix.ai,” he said. “And that’s what we think is the right approach to do it.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the “Agentic AI Unleashed: The Future of Digital & IT Operations” event:

(* Disclosure: TheCUBE is a paid media partner for the “Agentic AI Unleashed: The Future of Digital & IT Operations” event. Neither Fabrix.ai, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE