AgentGUI: New Interface for Observing and Steering AI Agents

Researchers have introduced AgentGUI, a locally hosted graphical user interface designed to observe and steer AI agents during complex, long-running tasks. The tool addresses the growing gap between autonomous agent capabilities and human oversight. AgentGUI offers rich trajectory visualizations, manual and automated steering options, and integration with both open-source and frontier agent frameworks. A controlled user study found that participants using AgentGUI identified key elements from agent traces 38% faster (p = 0.023). In a preliminary experiment, the automated drift prevention feature improved task completion rates for small local agents by up to 34 percentage points across a model ladder ranging from 0.8B to 9B parameters (N=50 runs per mode).