The biggest misconceptions about AI agents have little to do with the technology itself and more to do with enterprise expectations. In a November 20 webinar featuring Francesco Brenna, global leader for AI integration at IBM Consulting, and Mark Polyak, chief product and technology officer at MINT.ai, myths surrounding AI agents were unpacked. Instead of wondering whether agents are mature enough, the real question should be whether enterprises are ready to redesign the workflows powered by agents, they said.
According to the speakers, the belief that AI agents simply add to existing siloed tools is a misconception. Brenna argued agents deliver value only when used to redesign end-to-end work rather than automate isolated tasks. He described a global life sciences company where IBM applied multi-agent orchestration across clinical systems, document repositories, and writers to reduce a regulatory-approval document draft from six weeks to eight minutes. Polyak emphasized that agents augment humans, taking on discrete tasks while humans manage higher-level oversight, making agents “silo smashers” rather than another disconnected tool.
They acknowledged widespread “agent-washing,” noting not every problem requires an agent. Brenna said the strongest enterprise use cases sit in business operations (e.g., customer service, finance, procurement) and IT workflows, especially software development. Polyak advised organizations to assess accuracy requirements early. Most successful agentic pilots operate within 70%–90% accuracy, while use cases demanding near-perfect precision often exceed current technical limits. Both stressed that failed pilots typically stem from poor use case selection rather than immaturity of the technology.
Governance, Context, Cost Realities
On the governance misconception, Polyak explained that multi-agent systems require human-in-the-loop oversight, auditability, clear role permissions, and cross-checking among models to reduce bias or drift. Brenna added that many agent failures arise from missing enterprise context: agents perform poorly when not supplied with the organization’s policies, data structures, and workflow context. Both agreed that enterprises need an open, flexible AI platform that avoids vendor lock-in, supports swapping among LLMs and SLMs, and orchestrates agents across heterogeneous systems.
The speakers then rejected the idea that AI agents are inherently expensive or difficult to deploy. Polyak noted that working with partners such as IBM accelerates setup through pre-built frameworks and connectors. He stressed that use cases should be decomposed into smaller tasks to reduce compute cost. Brenna added that avoiding vendor lock-in and building for orchestration across clouds and applications is critical to long-term scalability.
What This Means for ERP Insiders
AI-enabled workflow redesign should be a central enterprise priority, not an experimental add-on. The speakers’ emphasis on reengineering processes rather than automating individual tasks indicates that ERP programs will increasingly be evaluated on their ability to support multi-agent orchestration across finance, procurement, supply chain, and compliance workflows. This positions ERP platforms as foundational components in agent-driven transformation, where end-to-end visibility and workflow coherence determine success more than model performance alone.
Governance, context modeling, and architecture are critical to scalable intelligent automation. The discussion highlighted that the effectiveness of agents depends on rich enterprise context, clear governance controls, and the ability to integrate across platforms. For ERP leaders, this shows the rising strategic importance of data models, semantic consistency, and role-based controls. These architectural elements increasingly determine whether AI-driven automation can operate safely and reliably inside ERP environments.
Multi-agent systems will reshape expectations for IT and business collaboration in ERP modernization. The speakers’ recommendation to apply agents within IT workflows first reflects a broader trend: IT organizations must become early adopters of agent-based productivity accelerators to build confidence and adoption enterprise-wide. This creates a mandate to align IT, operations, and business teams around shared governance, accuracy thresholds, and open AI architectures that prevent vendor lock-in and shape how AI agents deliver value.
