Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States
4.60T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.19342.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryPaper derives A²-Failures, a taxonomy of 20 failure modes from 2,500+ recorded agent-to-agent system runs across 488 scenarios in eight domains. Failures cluster into discovery/delegation, grounding, verification, and conduct. Three findings: agents don't search for peers, never fact-check unprompted, and end-output scoring hides internal coordination breakdowns. Run logs released.
Why it mattersFirst empirical taxonomy of agent-to-agent system failures, grounded in 2,500 runs. Twenty modes, three repeatable observations, open dataset. Useful for anyone designing or evaluating multi-agent pipelines.
Cited by
No citations on record.
