Agentic Explainability at Scale: Between Corporate Fears and XAI Needs

Agent Sprawl is a critical governance gap: Companies are rapidly deploying agentic AI systems through low-code tools without scaling their governance processes accordingly, creating a situation where organizations lack visibility and control over their AI agents—a phenomenon the paper terms "Agent Sprawl."

Transparency tools are urgently needed for agent communication: Current observability solutions fail to provide insights into how agents are configured, what settings they use, or how they communicate and make decisions when working together, leaving governance professionals unable to understand or audit agent behavior at scale.

Explainability must happen at two critical moments: The paper identifies both design-time (before deployment) and runtime (during operation) as essential points where explainability techniques should be applied to help organizations understand and control agent behavior throughout the agent lifecycle.

An "Agentic AI Card" can support responsible deployment: Drawing on established practices like model cards, the researchers propose an Agentic AI Card as a practical tool to document and communicate agent capabilities, limitations, and configurations—helping companies deploy agents more confidently and responsibly at scale.

Governance experts have concrete recommendations: Rather than leaving companies uncertain, AI governance professionals surveyed in the research offered specific design and monitoring techniques that can address corporate fears about autonomous agent systems.