IBM puts agentic workloads at the heart of its AI operating model

Agentic workloads are transforming how enterprises implement artificial intelligence. As AI systems increasingly take on ongoing responsibilities, organizations must manage decision-making and actions across their data, applications, and infrastructure. IBM has developed an AI operating model based on lessons learned from clients, identifying key requirements for this evolving landscape, according to Ritika Gunnar, IBM's chief performance marketing officer.

Gunnar highlighted that as AI systems become more effective, they continually drive workloads that must always be operational. IBM’s AI operating model consolidates four components that organizations may already possess from various providers: data, workloads, automation, and hybrid cloud. These elements must work together to effectively handle agentic workloads, with data serving as the foundation of enterprise knowledge and automation providing oversight and optimization.

Ensuring proper integration with existing systems, such as SAP and Salesforce, is crucial, as employees often rely on these established tools. Gunnar emphasized that data governance is essential, as it determines who has access to data and when, enabling real-time flow of information and context with provenance. Capturing tacit knowledge, which employees develop through experience, remains a challenge, as prompts and context can aid in conveying judgment but are not sufficient on their own.

Agent activity produces decision logs and prompt records, which, when combined, facilitate a feedback loop between enterprise data and AI workloads. Performance metrics, including accuracy, timeliness, and cost, guide evaluations of AI agents' effectiveness in achieving business outcomes. Overall, IBM's AI operating model calls for a comprehensive examination of both the work and the technology supporting it, focusing on business process transformation and ensuring optimal outcomes through continuous evaluation and improvement.