AgentGraph: Trace-to-Graph Platform for Interactive Analysis and Robustness Testing in Agentic AI Systems
Association for Artificial Intelligence 2026, Zekun Wu, Maria Perez-Ortiz, Sahan Bulathwela, Seonglae Cho, Emre Kazim, Theo King, Adriano Koshiyama, Umar Mohammed
Open MIND
Problems Identified (4)
Agentic Execution Interpretability: Multi-step agentic AI execution patterns are difficult to interpret.
Manual Trace Reconstruction: Existing observability platforms require manual inspection of traces to reconstruct system structure and reasoning.
Agentic Execution Interpretability: Multi-step agentic AI execution patterns are difficult to interpret.
Manual Trace Reconstruction: Existing observability platforms require manual inspection of traces to reconstruct system structure and reasoning.
Proposed Solutions (5)
Trace-to-Knowledge-Graph Platform: AgentGraph converts execution logs into interactive knowledge graphs and actionable insights.
Trace-Linked Agent Graph Representation: AgentGraph represents agents, tasks, tools, data inputs and outputs, and humans as nodes with typed edges for execution relations, linking graph elements to exact trace spans.
Trace-Grounded Robustness Analysis: AgentGraph uses the graph representation to support qualitative failure detection and optimization recommendations plus quantitative robustness evaluation using perturbation testing and causal attribution.
Trace-to-Knowledge-Graph Platform: AgentGraph converts execution logs into interactive knowledge graphs and actionable insights.
Trace-Linked Agent Graph Representation: AgentGraph represents agents, tasks, tools, data inputs and outputs, and humans as nodes with typed edges for execution relations, linking graph elements to exact trace spans.
Results (3)
Verifiable Trace Analysis:
Failure Detection And Optimization Support:
Robustness Evaluation Support:
Research Domain
Agentic AI observability, trace analysis, and robustness testing