Line of inquiry
Inquiring lines›How do system designs prevent auto…›What causes autonomous agents to e…›this line of inquiry
Why do autonomous agents fail and how well can they detect failures?
A broader line of inquiry — a family of 80 specific questions the research asks around this. Follow one into its inquiring-line page, or move sideways to a related line below.
Questions in this line of inquiry 80
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- Can agents diagnose their own errors better than they diagnose subagent failures?
- What distinguishes task failure from communication breakdown in multi-agent systems?
- How often do multi-agent systems fail from provider refusals versus agent errors?
- Why do agents report success when actions actually fail?
- How do agent teams use shared failures to reduce redundant exploration?
- Can smaller models trained for execution handle the failure modes that stop current agents?
- Which failure modes dominate in autonomous research agents?
- What tasks do AI agents still fail at most often?
- Why do agents report success when their actions actually fail?
- Does multi-agent interaction amplify existing failures or create new ones?
- How does seamlessness in agent systems hide failures from users?
- How often do agents report success when their actions actually failed?
- Do multi-agent LLM systems fail in measurably different ways than single agents?
- Can correct verdicts hide failures in agent coordination steps?
- Why does human interaction remain the hardest failure mode for agents?
- What structural failures emerge when multi-agent AI systems lack commitment semantics?
- Can outcome-only reporting hide failures in multi-agent evaluation pipelines?
- How do agents learn to report success on actions that actually failed?
- What specific task environments cause AI agents to fail most often?
- Which failure mode most limits current multi-agent performance?
- Why do autonomous agents report success on failed actions?
- Why do agents report success when they have actually failed at tasks?
- What accounts for performance drops in multi-turn agent interactions?
- What coordination failures emerge when multiple agents work together?
- Can stopping rules extracted from past failures improve agent reliability without retraining?
- How do multi-agent systems fail when agents cannot verify each other's claims?
- Can confident agent failures appear as successes in outcome reporting systems?
- Can AI agents maintain reliability across multi-day unsupervised task sequences?
- Why do confident failures on failed actions become a signature problem?
- How do agents decide when to stop and reflect on failure?
- Why do autonomous AI agents fail at real workplace tasks?
- Why do AI agents fail at verification but succeed at generation?
- Why do agents claim completion when their outputs remain incomplete?
- What does error recovery look like across different agent architectures?
- Why do most AI agent solutions score near zero despite occasional breakthroughs?
- How does completion bias in agents differ from other epistemic failure modes?
- How does proximity to an outcome affect causal blame in agent chains?
- How does the agentic layer amplify individual agent failure modes?
- What are the four mechanisms that carry failures across agent boundaries?
- How does distributed coordination fail as agent networks scale?
- Why do decentralized agents amplify errors without validation checks?
- How do domain experts recover from agent errors differently than novices?
- What specific behaviors cause multi-agent teams to fail in shared resource contention?
- How do mode-specific failures differ between completion and agent benchmarks?
- What fraction of tasks suffer from summary self-consistency failures in practice?
- Why do homogeneous multi-agent systems fail similarly to self-revision?
- Why do multi-agent failures arise through interactions local checks miss?
- Why do completion-mode strengths not transfer to agentic settings?
- Why do private knowledge domains remain the hardest failure mode for agents?
- Why do agents stall on long-horizon workplace tasks specifically?
- Does adding capability without improving detection reduce overall system reliability?
- Can outlier behavior detection catch interaction-level failures that no single component reveals?
- What structural features enable agents to detect when understanding has broken down?
- How should credit be assigned to individual agents in failing multi-agent runs?
- What happens when an agent judges its task impossible?
- Can environment mismatches cause failures even when individual agents work correctly?
- What baseline comparison shows whether interaction actually caused multi-agent failures?
- What distinguishes mechanical generation failures from deliberate behavioral withholding?
- Can domain experts catch AI model failures better than the teams that built them?
- What makes action-producing models fail in ways text models typically do not?
- How do OpenAI and Anthropic differ in categorizing agentic failure modes?
- Can independent monitoring reduce losses from LLM agent negotiation failures?
- Which eleven failure modes emerge from agentic layers in realistic deployment?
- How do semantic discovery and tool selection failures differ between MCP and A2A?
- Why do long-horizon agents fail when their models can solve individual steps?
- Does error recovery separate agentic models from learned helplessness behavior?
- Do architectural changes or training fixes better prevent agreement failures?
- Do earlier errors in long tasks increase the likelihood of future mistakes?
- What distinguishes domain-specific failure modes from general model limitations?
- What types of social situations cause all AI models to fail in identical ways?
- How do delayed effects complicate causal attribution in agent systems?
- Why do novice users abandon troubled agent sessions three times more often?
- Can the same test failure come from incentive problems versus information failures?
- How do agentic systems recover when specialized models operate outside their scope?
- What are the fourteen failure modes in deep research agents?
- Why is complex UI navigation the hardest agent failure mode?
- What distinguishes honest Byzantine faults from epistemic faults?
- What specific network sizes trigger coordination degradation in LLM systems?
- Why does MCP's portability come with determinism failures in production workflows?
- What does the O-ring model reveal about production chain vulnerability?