Agentic explainability techniques address enterprise governance gaps
Yomna Elsayed and Cecily Jones propose design-time and runtime explainability techniques and an Agentic AI Card prototype to help enterprises govern autonomous agents at scale while managing risks from "Agent Sprawl."
Score breakdown
Developers and governance teams deploying autonomous agents can use design-time and runtime explainability techniques plus the Agentic AI Card framework to maintain visibility and control over agent behavior as adoption scales, reducing deployment risk.
- 01Agent Sprawl occurs when enterprises scale agentic AI adoption through low-code applications without corresponding increases in governance processes and expertise
- 02Existing shadow AI tools can identify agents but lack visibility into agent configuration, settings, and decision-making during agent-to-agent communication and orchestration
- 03The paper proposes design-time and runtime explainability techniques informed by AI governance experts to address enterprise governance concerns
Enterprise adoption of agentic AI is accelerating, but governance infrastructure is not keeping pace. As companies deploy autonomous agents through low-code applications, a gap emerges between the speed of agent proliferation and the maturity of governance processes and expertise. This creates "Agent Sprawl"—a state where agents operate with insufficient oversight and control. While existing shadow AI tools can help discover and identify agents in use, they provide limited insight into agent configuration, settings, or the decision-making processes that occur during agent-to-agent communication and orchestration.
Yomna Elsayed and Cecily Jones investigate the specific concerns of AI governance professionals in enterprise environments and propose targeted solutions.
Yomna Elsayed and Cecily Jones investigate the specific concerns of AI governance professionals in enterprise environments and propose targeted solutions. Drawing on input from AI governance experts, they recommend both design-time explainability techniques (applied during agent development and deployment) and runtime explainability techniques (applied during agent execution). The paper culminates in a preliminary prototype of an Agentic AI Card—a structured artifact intended to document and communicate agent behavior, configuration, and decision-making in a way that helps enterprises manage risk and maintain confidence as they scale agentic AI deployments.
Key facts
- 01Agent Sprawl occurs when enterprises scale agentic AI adoption through low-code applications without corresponding increases in governance processes and expertise
- 02Existing shadow AI tools can identify agents but lack visibility into agent configuration, settings, and decision-making during agent-to-agent communication and orchestration
- 03The paper proposes design-time and runtime explainability techniques informed by AI governance experts to address enterprise governance concerns
- 04An Agentic AI Card prototype is introduced as a tool to help companies deploy agents at scale with greater transparency and control