IBM combines agent workloads with the needs of always-on AI

Agentic workloads represent a new operational challenge for companies integrating artificial intelligence into production. As autonomous systems assume ongoing responsibility, companies must support decisions and actions in their data, applications and infrastructure.

IBM Corp.’s approach based on the experiences of customers scaling AI in their businesses. According to Ritika Gunnar (pictured), IBM’s chief performance marketing officer, this work helped identify the requirements underlying the AI ​​operating model.

“As you become more successful in using AI, particularly when AI takes action, you now have workloads that are continually used,” she said. “You have the incentive to always be active because these agents are always working for you.”

Gunnar spoke with Dave Vellante and John Furrier of theCUBE Research for IBM’s AI Operating Model interview series in an exclusive interview on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed enterprise integration, data management, human oversight, and evaluating AI against business outcomes in preparation for IBM TechXchange 2026. (*Disclosure below.)

Connect agent workloads across the enterprise

IBM’s AI operating model combines four components that companies may already source from different vendors. Applying them means considering established applications and business processes alongside emerging AI systems, Gunnar explained.

“You need to be able to integrate with your existing systems,” she said. “People aren’t going to run away from their existing SAP systems or their existing Salesforce systems. What does it mean to integrate with those and have workloads that are representative of your past and your shared future?”

The components have different roles: data provides up-to-date business knowledge, workloads include applications and workflows, automation supports monitoring and optimization, and the hybrid cloud supports distributed operations. Their effectiveness depends on how they work together across agent workloads, according to Gunnar.

“If you have your data and that data is feeding your systems, but you don’t really care about the automation so that you have complete observability of all your workloads with that AI, then you don’t necessarily have the optimal opportunity to select the right model or the right agents for those types of workloads,” she said.

Incorporate governance and human judgment into decisions

The information that supports an agent’s decision can include operating systems, edge devices, and context generated by other agents. To provide a regulated context, companies must consider both the completeness of this information and the permissions associated with it, Gunnar explained.

“We believe that if you can take data and shift a lot of the governance to the left – a lot of who has the right access to the right data at the right time – it becomes important because you’re sharing that data in real time with the context and provenance of who actually owns that data,” she said. “As a result, agents make much better decisions.”

Companies also rely on employees to acquire knowledge through experience rather than explicitly recording it in their systems. Prompts and other context can help convey this judgment, but grasping it remains an ongoing challenge, Gunnar emphasized.

“When you think about the workload today … there’s a lot of it that’s going to be autonomous in nature,” she said. “But in most cases, when it comes to your essential systems, there are people on the cutting edge who are able to bring some of that tacit knowledge to where AI is not yet mature enough to even get there.”

Evaluate performance against business results

Agent activities generate logs of decisions and actions, while prompts record the instructions communicated to these systems. Together, this information supports a continuous feedback loop between enterprise data and AI workloads, Gunnar explained.

“You can actually have several different permutations where this agent makes a decision, and a different set of evaluation loops that actually do an evaluation: ‘Did this agent actually make the decisions?'” she said. “Would there have been a more powerful, more accurate and more cost-effective way to achieve this?”

Performance includes accuracy, timeliness and cost, with the intended business outcome as the guide for evaluating these measures. Therefore, according to Gunnar, to integrate AI into production, companies must examine the work itself and the technology that supports it.

“Make sure you have all of those elements and that you’re really thinking about the outcomes and ready to start looking at what it means to take a look at business process transformation,” she said. “That’s really where it’s at.”

Here is the full video interview, part of SiliconANGLE and theCUBE’s coverage of IBM’s AI Operating Model interview series:

(*Disclosure: TheCUBE is a paid media partner for IBM’s AI Operating Model interview series. Neither IBM, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over theCUBE or SiliconANGLE content.)

Photo: SiliconANGLE