arXiv cs.AI· Shravika Mittal, Sunnie S. Y. Kim, Scott Counts, Jennifer Neville, Debarati Das, Philippe Laban·· 3 小时前AI 评分42
看护 AI 智能体：MANDATE 研究提出智能体委托失败的以用户为中心分类法
Babysitting AI Agents: A User-Centered Taxonomy and Study of Agentic Delegation Failures
AI 导读
研究者基于 163,487 条 Reddit 帖子构建了 MANDATE 分类法，将 AI 智能体委托失败归纳为 7 组共 23 个类别。随后对 73 名智能体用户的后续研究发现，被认为越严重的失败越难被察觉，且有亲身经历的用户更容易注意到失败。研究还指出部分常见失败（如智能体运行却无进展）缺乏研究关注。
正文 · 原文
Abstract:Users are increasingly delegating tasks to AI agents, allowing agents to perform long-running, multi-step workflows on their behalf. However, agents may perform these tasks in ways that depart from users' expectations for how they should be carried out, leading to what we call agentic delegation failures. We leverage a large corpus of 163,487 Reddit posts, describing AI agent use, to develop MANDATE: a taxonomy of 23 failure categories organized into seven groups. A follow-up study with 73 AI agent users examined the relationship between prevalence and perceptions of severity and noticeability, finding that failures rated as more severe are considered harder to notice, and that prior experience with a failure is associated with greater noticeability. Finally, we assess the maturity of research on each failure and identify commonly encountered failures that have received little attention (e.g., agents running without making progress). Our findings inform opportunities to make delegation easier for users to oversee and better studied by the research community.
| Subjects: | Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.34748 [cs.HC] |
| (or arXiv:2610.34748v1 [cs.HC] for this version) | |
| https://doi.org/10.48550/arXiv.2610.34748 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shravika Mittal [view email]
[v1]
Mon, 26 Oct 2026 02:49:17 UTC (8,240 KB)
来源：arXiv cs.AI · arxiv.org