Dhanesh Ramachandram

Position: Accountable Deployment of Agentic AI Demands Layered, System-Level Interpretability

Agentic AI systems behave through trajectories: they plan, invoke tools, update memory, and coordinate over multiple steps. However, interpretability remains largely model-centric, focused on explaining single predictions rather than tracing long-horizon behavior and responsibility across interactin…