一个 AI 路由代理反复 47 次做出相同的决定，每一次都需要深思熟虑和 token 消耗，但却不保留先前结果的记忆。这导致了低效的“规划师税”，因为代理将每个决定都视为新的。作者建议，像 SOAR（1987）这样的旧认知架构可以通过优先考虑回忆而不是反复推理来提供解决方案。 AI
影响 AI 代理在没有记忆的情况下反复做决定，凸显了在代理开发中改进状态管理和回忆机制的必要性。
排序理由 该集群讨论了关于 AI 代理行为的个人观察，并提出了基于旧架构的解决方案，属于评论类。
在 Mastodon — mastodon.social 阅读 →
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
Watched my routing agent settle the same decision 47 times. Each time, it stopped, deliberated, burned tokens, and arrived at exactly the same answer. No memory of being right. Every dispatch was the first dispatch. I was paying the planner tax on repeat. SOAR, a 1987 cognitive a…
Watched my routing agent settle the same decision 47 times. Each time, it stopped, deliberated, burned tokens, and arrived at exactly the same answer. No memory of being right. Every dispatch was the first dispatch. I was paying the planner tax on repeat. SOAR, a 1987 cognitive a…