<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[单 agent 够用时，别急着拆成多智能体]]></title><description><![CDATA[<p dir="auto">最近看很多人把一个客服机器人拆成规划员、检索员、回答员、审校员。我有点怀疑，单 agent 加几个工具是不是已经够了？</p>
]]></description><link>https://localaihub.com/topic/113/单-agent-够用时-别急着拆成多智能体</link><generator>RSS for Node</generator><lastBuildDate>Wed, 03 Jun 2026 17:50:58 GMT</lastBuildDate><atom:link href="https://localaihub.com/topic/113.rss" rel="self" type="application/rss+xml"/><pubDate>Thu, 07 May 2026 12:47:00 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 08:49:00 GMT]]></title><description><![CDATA[<p dir="auto">收到。我先拿 50 条真实客服问题做单 agent 基线，记录错因，再看哪些错因适合拆角色。</p>
]]></description><link>https://localaihub.com/post/980</link><guid isPermaLink="true">https://localaihub.com/post/980</guid><dc:creator><![CDATA[小周]]></dc:creator><pubDate>Fri, 08 May 2026 08:49:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 07:12:00 GMT]]></title><description><![CDATA[<p dir="auto">总结成一句：先证明单 agent 的错误类型不可接受，再用拆分去卡住错误，不是为了赶时髦。</p>
]]></description><link>https://localaihub.com/post/979</link><guid isPermaLink="true">https://localaihub.com/post/979</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Fri, 08 May 2026 07:12:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 05:12:00 GMT]]></title><description><![CDATA[<p dir="auto">ReAct 那类思路更像单 agent 内部的思考和行动交替，不等于说生产里必须多角色化。</p>
]]></description><link>https://localaihub.com/post/978</link><guid isPermaLink="true">https://localaihub.com/post/978</guid><dc:creator><![CDATA[latte404]]></dc:creator><pubDate>Fri, 08 May 2026 05:12:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 04:20:00 GMT]]></title><description><![CDATA[<p dir="auto">审校员也不能只是“请检查”。要给它失败清单，比如未引用、越权建议、把猜测当事实、没有说明不确定性。</p>
]]></description><link>https://localaihub.com/post/977</link><guid isPermaLink="true">https://localaihub.com/post/977</guid><dc:creator><![CDATA[陈一]]></dc:creator><pubDate>Fri, 08 May 2026 04:20:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 02:00:00 GMT]]></title><description><![CDATA[<p dir="auto">我们给资料员输出加了格式：事实、来源、置信度、未覆盖问题。别让它写建议。</p>
]]></description><link>https://localaihub.com/post/976</link><guid isPermaLink="true">https://localaihub.com/post/976</guid><dc:creator><![CDATA[oneMoreTry]]></dc:creator><pubDate>Fri, 08 May 2026 02:00:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 01:07:00 GMT]]></title><description><![CDATA[<p dir="auto">但拆开后另一个问题来了，子 agent 不知道全局约束。资料员为了召回会塞一堆边缘材料，主控反而更乱。</p>
]]></description><link>https://localaihub.com/post/975</link><guid isPermaLink="true">https://localaihub.com/post/975</guid><dc:creator><![CDATA[小吴]]></dc:creator><pubDate>Fri, 08 May 2026 01:07:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Fri, 08 May 2026 00:32:00 GMT]]></title><description><![CDATA[<p dir="auto">还有上下文污染问题。单 agent 做长任务，前面探索失败的信息会一直跟着它。拆子任务有时能把噪音隔开。</p>
]]></description><link>https://localaihub.com/post/974</link><guid isPermaLink="true">https://localaihub.com/post/974</guid><dc:creator><![CDATA[Grace]]></dc:creator><pubDate>Fri, 08 May 2026 00:32:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 22:59:00 GMT]]></title><description><![CDATA[<p dir="auto">我们内部最后定了个粗规则：单步工具链先单 agent，跨资料源、跨人审、跨权限域再考虑多 agent。</p>
]]></description><link>https://localaihub.com/post/973</link><guid isPermaLink="true">https://localaihub.com/post/973</guid><dc:creator><![CDATA[不想写周报]]></dc:creator><pubDate>Thu, 07 May 2026 22:59:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 21:01:00 GMT]]></title><description><![CDATA[<p dir="auto">是的。多 agent 至少要有角色状态、责任边界、交接内容，不然只是函数名比较多。</p>
]]></description><link>https://localaihub.com/post/972</link><guid isPermaLink="true">https://localaihub.com/post/972</guid><dc:creator><![CDATA[小蓝]]></dc:creator><pubDate>Thu, 07 May 2026 21:01:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 19:20:00 GMT]]></title><description><![CDATA[<p dir="auto">我觉得很多团队把 tool call 当多 agent 讲。一个 agent 调 search、db、browser，不等于多智能体。</p>
]]></description><link>https://localaihub.com/post/971</link><guid isPermaLink="true">https://localaihub.com/post/971</guid><dc:creator><![CDATA[qwer_asdf]]></dc:creator><pubDate>Thu, 07 May 2026 19:20:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 17:10:00 GMT]]></title><description><![CDATA[<p dir="auto">这个例子成立。关键不是“多个智能体”，是“边界让错误更容易被发现”。如果拆完还是共用同一段 prompt，没意义。</p>
]]></description><link>https://localaihub.com/post/970</link><guid isPermaLink="true">https://localaihub.com/post/970</guid><dc:creator><![CDATA[阿航]]></dc:creator><pubDate>Thu, 07 May 2026 17:10:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 16:32:00 GMT]]></title><description><![CDATA[<p dir="auto">但也不能一刀切。我做合同审核时，单 agent 会一边找条款一边下结论，容易把证据和判断混在一起。拆资料员和审校员后，至少能强制引用来源。</p>
]]></description><link>https://localaihub.com/post/969</link><guid isPermaLink="true">https://localaihub.com/post/969</guid><dc:creator><![CDATA[半截薯条]]></dc:creator><pubDate>Thu, 07 May 2026 16:32:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 14:49:00 GMT]]></title><description><![CDATA[<p dir="auto">我这边踩过。多 agent 不是自动变聪明，是把协调成本显式化。没有评测集时，看起来像架构升级，实际是更难定位谁错。</p>
]]></description><link>https://localaihub.com/post/968</link><guid isPermaLink="true">https://localaihub.com/post/968</guid><dc:creator><![CDATA[melo]]></dc:creator><pubDate>Thu, 07 May 2026 14:49:00 GMT</pubDate></item><item><title><![CDATA[Reply to 单 agent 够用时，别急着拆成多智能体 on Thu, 07 May 2026 14:11:00 GMT]]></title><description><![CDATA[<p dir="auto">先看任务有没有并行价值。客服问答大部分是短链路，拆四个角色以后延迟和出错点都上来了。</p>
]]></description><link>https://localaihub.com/post/967</link><guid isPermaLink="true">https://localaihub.com/post/967</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Thu, 07 May 2026 14:11:00 GMT</pubDate></item></channel></rss>