Survey on LLM Agent Failure Taxonomies
dair_ai · x · 2026-07-08
A review from the University of Oxford analyzes 27 papers on LLM agent benchmarks, taxonomies, and audits across 19 benchmarks, aiming to establish a unified cross-evaluation failure taxonomy.
The paper identifies 6 common failure categories: tool calling and parameter errors, planning and constraint satisfaction failures, and long-range degradation caused by accumulated long contexts. It notes that many agents repeatedly exhibit similar issues across different evaluations.
More from coding & agent
- A 9B Ollama agent can run a fully local DJ radio with tools, memory, and TTS — pinku1 · 2026-07-27
- Bugbot rejects an MCP permission flag because it would break path-scoped isolation — zeeg · 2026-07-27
- One GPT-5.6 agent is guarding a Blink security system while another makes a parody rap album — repligate · 2026-07-27
- An agent got unblocked by reusing a logged-in browser, not stealth tricks — armanidev_ · 2026-07-27
- Paper argues graph topology can become the core operating system for AI agents — theomitsa · 2026-07-27
- Claude Code desktop adds UI markup feedback for smoother visual editing — EricBuess · 2026-07-27