Llm agents fail early with unnoticed errors in long tasks

Dependable Long-Horizon LLM Agents: A Trajectory-Centered Empirical Study of Where Errors Enter and Which Controls Reach Them

Artificial Intelligence

Summary

This paper studies why AI systems based on large language models often fail during long tasks. The authors found that errors usually start early, often due to wrong decisions or made-up information, and these errors often go unnoticed or uncorrected despite evidence in the AI's own process. They built a detailed way to trace where and how errors appear and spread, showing many failures could have been caught or fixed earlier. Their analysis also revealed a kind of failure not described by usual methods, involving crashes after content filtering.

What this means in practice

  • For ai system engineers: Diagnose and debug long-task failures by tracing error origins and propagation in multi-agent LLM systems using the provided annotation protocol.
  • For software reliability teams: Improve failure detection and response strategies in AI workflows by identifying earlier fault signs and unused evidence in execution trajectories.

Authors

Shaohuang Wang, Jinghao Wen, Xulei Zhang, Xun Jiao

Abstract

When an LLM agent fails part-way through a long task, the usual analysis records a symptom: a failure-mode label, or the agent and step judged responsible. It does not say where the error entered, which state carried it to the symptom, whether anything in the trajectory contradicted it, or which control could have caught it. We define an episode record that ties a candidate fault to its origin, propagation, detection, response and consequence, located by operational phase and state domain, and that names the control families whose observation and reach cover it. We turned the record into a written annotation protocol and applied it to 95 failed runs of six multi-agent systems from the public Who&When and MAD corpora, reading each trajectory in full. Faults enter early (median at 0.18 of the run), mostly as wrong decisions or fabrications over task state, and in 37 of 88 Who&When runs before the step the corpus' human labels mark as decisive. In 72% of failed runs the trajectory already contained evidence against the error before the failure; in 88% of those no agent acted on it. Something flagged a deviation in 31% of runs, but no response removed the deviated state. Independent judges reading the unused evidence, task contracts fixed at admission, and re-checking beliefs against the environment reach most episodes; restoring the agent's own state would have sufficed in 35 of 95 runs. A text scan of the raw corpus also shows that 45% of the hand-crafted Who&When failures ended with a provider content-filter crash after which the harness emitted a final answer no agent produced, a failure class that neither symptom taxonomies nor responsible-agent labels represent. Protocol, records and scripts are included.