Tool Calling is Not Just Plumbing for AI Agents

In this engaging and technically rich session from the Artificial Unintelligence Conference 2025, Roy Derks (IBM) breaks down one of the most misunderstood aspects of modern AI development: tool calling. With his trademark clarity and humor, Roy explains how tool calling and the emerging Model Context Protocol (MCP) form the backbone of multi-agent systems — enabling AI models to interact with real-world data, applications, and APIs.

Far from being “just plumbing,” tool calling defines how intelligent agents actually reason, act, and deliver meaningful outcomes. Roy shares practical insights from the field, illustrating how abstraction, context management, and human oversight come together to build scalable, reliable, and secure AI architectures.

Key themes include:

Why tool calling is foundational to agentic intelligence

How the Model Context Protocol (MCP) enables tool interoperability

The risks of poor tooling and fragile integrations

Why abstraction is key to safe, scalable agent design

The continuing importance of human iteration and oversight