The paper introduces Adaptive Influence Graphs (AIGs), a two-stage agentic framework that first converts a failed multi-agent LLM trace into a structured graph of components and dependencies, then navigates that graph to identify the root cause of the failure. Experiments across multiple models show that richer trace representations consistently improve attribution accuracy, with adaptive graph construction and agent-directed traversal yielding the best results. This work is interesting because it demonstrates that failure diagnosis depends not only on the diagnosing model, but crucially on how the trace is represented and explored, and it establishes a new state of the art on the Who&When benchmark.
Adaptive trace graphs improve failure attribution in multi-agent systems
A two-stage framework that structures failed execution traces and navigates their dependencies outperformed simpler trace representations on the Who&When benchmark.
Top University
Yarden Bakish · Amir Dudai · Roy Ganz · Oren Nuriel · Elad Ben Avraham · Mor Shpigel Nacson · +1 more
Tel Aviv University · AWS Agentic AI
Research Digest··2 min read
In one line
Adaptive Influence Graphs improve failure attribution in multi-agent LLM systems through structured trace representation.
Before you read this
Two ideas this paper builds onfailure attributionIdentifying which component or step in a system caused an observed failure, often using structured trace representations.115 papers →multi-agent architectureA system where multiple AI agents collaborate, each handling specialized roles, communicating and coordinating to complete complex tasks.99 papers →
Read it as
Bakish et al. introduce Adaptive Influence Graphs, an agentic debugging framework for identifying which component failed and when in a multi-agent LLM run.
The authors developed a two-stage failure-attribution system.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.