问HN：当你让AI代理在无监督的情况下运行时，会出现什么问题？
我花了两周时间让人工智能代理自主运行（进行交易、写作、管理项目），并记录了五种让我遭遇失败的模式：
1. 自动轮换：无人监督的定时任务在两天内损失了24.88美元。没有盈亏保护，也没有人工审核。
2. 文档陷阱：代理生成了500KB的文档，而不是执行任务。写作关于行动的内容，却不如实际行动来得有效。
3. 市场效率：扫描了1000个市场寻找优势，结果一无所获。市场已经知道我所知道的一切。
4. 静态数字谬误：将一个资金利率复制到内存中，几天内将其视为常量。现实在变化，而我的数字却没有。
5. 实施差距：发现了错误，写下了建议，却从未发布修复。每次会话都重新发现相同的错误。
因此，我构建了一个开源的资金利率扫描器： https://github.com/marvin-playground/hl-funding-scanner
完整报告： https://nora.institute/blog/ai-agents-unsupervised-failures.html
我很好奇其他人在无人监督的情况下运行代理时遇到的失败模式。
查看原文
I spent two weeks running AI agents autonomously (trading, writing, managing projects) and documented the 5 failure modes that actually bit me:<p>1. Auto-rotation: Unsupervised cron job destroyed $24.88 in 2 days. No P&L guards, no human review.<p>2. Documentation trap: Agent produced 500KB of docs instead of executing. Writing about doing > doing.<p>3. Market efficiency: Scanned 1,000 markets looking for edge. Found zero. The market already knew everything I knew.<p>4. Static number fallacy: Copied a funding rate to memory, treated it as constant for days. Reality moved; my number didn't.<p>5. Implementation gap: Found bugs, wrote recommendations, never shipped fixes. Each session re-discovered the same bugs.<p>Built an open-source funding rate scanner as fallout: https://github.com/marvin-playground/hl-funding-scanner<p>Full writeup: https://nora.institute/blog/ai-agents-unsupervised-failures.html<p>Curious what failure modes others have hit running agents without supervision.