Anthropic 称 1 AI agent击败了许多子剂，现在模型可以单独处理复杂问题。
关键点
- Anthropic建议公司评价它们是否能够将子agents合并成一个按需加载上下文的单agent.
- 公司的经验显示，许多专业的子agents可以导致更高的耐久性，成本上升，每次交接时信息丢失。
- Anthropic在GitHub上发布了一个迁移工具包，帮助团队实现过渡。
Anthropic发现，单张agent的点播上下文往往在质量，成本，和耐久性上超过多张agent的设置。 公司正在发布一个迁移工具箱，以帮助团队进行评估和切换。
Anthropic建议各公司评价它们是否能够合并子agents，因为一个单一的初级agent根据需要加载上下文，往往比一支专门化的车队产生更好的效果。 从Opus 4.5开始的较新的模型，连同Opus 5.5和Fable 5.1，可以直接在一个循环中管理复杂的计划，这也可以降低成本和等待时间。 在一次比较中，Anthropic 称的单agent架构运行速度约为多agent设置速度的两倍。
该建议来自多个企业部署的亲身体验，Anthropic的应用AI团队发现，一个使用所谓技能的agent经常在质量上同时击败单质的即时设计和子agent设置。 例如，一家公司在客户面前的agent内部有十几个专业的子agents，这使得难以保持一致的品牌经验，以高的耐久性损害可用性。
模型现在可以做很多管弦乐工作了
Anthropic认为，许多agent的设计遵循了熟悉的意向检测模式，固定处理管道，用户分割，单独推荐分支，以及坐在顶部的管弦乐器。 大部分工作是这个模型现在可以自己处理，供各自使用。
Anthropic称其为agents的"康威定律". 该架构成为Org图的复制品，每支队伍运输自己的子agent，直到数百个重叠的agents堆积起来，没有共享标准。 接下来是Anthropic所称的"弗兰肯普伦普特"，每支球队新增规则到系统时都会提示最终相互矛盾。 据该公司称，一个清洁的系统即时、四、五个精益数据工具以及固定的输出护栏覆盖了各自的大部分使用案例。
例如，Anthropic将Claude Code系统用于更新的模型的速率削减了80%以上，部分原因是技术和工具现在只在需要的时候才加载。 较新的模型也对自己的进步进行评论，检查自己的作品，这可以使得一些额外的控制层变得没有必要。
Anthropic 称子agents在单个agent的上下文窗口显示可测量性能退化，多个任务需要与单独的上下文windows并行运行，或者单个agent与工具选择发生斗争时仍然有意义。
为了一步一步的迁移，Anthropic在GitHub上发布了一个迁移工具包，让团队通过一个工作流程来识别现有模式，评价替代品，并验证最佳方法。 Anthropic 警告反对盲目合并，因为在一次测试中，有支持的agent，在子agent中折叠确实减少了延迟，但打破了之前的工作过程。 初级agent开始自己回答，而不是咨询合适的专家。
团队应防止技能库扩展
Anthropic不给每个团队自己的agent，而是通过工具搜索将包装工作流程推荐成技能和按需加载工具。 中央小队应拥有迅速和总体质量的系统。 另外，该模型工具选择的Anthropic 指出往往会降解到超过30到50个工具定义，因此每个团队中指定的所有者应当维护技能库，以阻止他们气球。
对于指南，产品规格，或计量定义等参考材料，Anthropic推荐知识基础，只有需要时，agent通过搜索工具查询。 这种信息也可以作为参考文件嵌入到技能中。 Anthropic建议不要通过代码获取上下文并直接将其注入提示。
Anthropic的调查结果与近期的研究一致。 瓦尔斯的一项研究发现，agent团队可以比单一的agents更快，但不能提供更高的质量。 根据研究，从10到30或100agents的扩展没有带来质量收益，但成本要高得多。 OpenAI开发商Eric Provencher也质疑agent群的价值，称它们"几乎肯定"浪费的钱。
AI News Without the Hype – 由人类治愈
订阅THE DECODER免费阅读，每周AI通讯，我们独家的"AI雷达"前沿报告每年6次，全面存档访问，访问我们的评论部分。
现在订阅
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Anthropic says one AI agent beats many subagents now that models can handle complexity alone
Key Points
- Anthropic recommends that companies evaluate whether they can consolidate sub-agents into a single agent that loads context on demand.
- The company's experience shows that many specialized sub-agents can lead to higher latency, rising costs, and information loss at every handoff.
- Anthropic has published a migration kit on GitHub to help teams make the transition.
Anthropic found that a single agent with on-demand context often outperforms multi-agent setups in quality, cost, and latency. The company is releasing a migration kit to help teams evaluate and make the switch.
Anthropic is recommending that companies evaluate whether they can consolidate sub-agents, because a single primary agent that loads context as needed often delivers better results than a fleet of specialized ones. Newer models starting with Opus 4.5, along with Opus 5.5 and Fable 5.1, can manage complex plans directly in a single loop, which can also reduce costs and wait times. In one comparison, Anthropic says the single-agent architecture ran about twice as fast as the multi-agent setup.
The recommendation comes from hands-on experience across multiple enterprise deployments, where Anthropic's Applied AI team found that a single agent using so-called skills frequently beat both monolithic prompt designs and sub-agent setups on quality. One company, for example, had a dozen specialized sub-agents inside its customer-facing agent, which made it hard to keep a consistent brand experience and hurt usability with high latency.
The model can now do much of the orchestration work itself
According to Anthropic, many agent designs follow a familiar pattern with intent detection, a fixed processing pipeline, user segmentation, separate recommendation branches, and an orchestrator sitting on top. Much of that is work the model can now handle on its own for the respective use case.
Anthropic calls it "Conway's Law" for agents. The architecture becomes a copy of the org chart, with each team shipping its own sub-agent until hundreds of overlapping agents pile up with no shared standards. What follows is what Anthropic calls a "Frankenprompt," where every team adds rules to the system prompt that end up contradicting each other. A clean system prompt, four or five lean data tools, and fixed output guardrails cover much of the respective use case, according to the company.
As one example, Anthropic cut the Claude Code system prompt for newer models by over 80 percent, partly because skills and tools now load only when needed. Newer models also comment on their own progress and check their own work, which can make some of the extra control layers unnecessary.
Anthropic says sub-agents still make sense when a single agent's context window shows measurable performance degradation, when multiple tasks need to run in parallel with separate context windows, or when a single agent struggles with tool selection.
For a step-by-step migration, Anthropic has published a migration kit on GitHub that walks teams through a workflow to identify existing patterns, evaluate alternatives, and verify the best approach. Anthropic warns against merging blindly, since in one test with a support agent, folding in a sub-agent did reduce latency but broke a previously working process. The primary agent started answering on its own instead of consulting the right specialist.
Teams should keep skill libraries from sprawling
Instead of giving every team its own agent, Anthropic recommends packaging workflows into skills and loading tools on demand through a tool search. A small central team should own the system prompt and overall quality. Separately, Anthropic notes that the model's tool selection tends to degrade beyond 30 to 50 tool definitions, so designated owners on each team should maintain the skill libraries to keep them from ballooning.
For reference material like guidelines, product specs, or metric definitions, Anthropic recommends knowledge bases that the agent queries only when needed through a search tool. That kind of information can also be embedded as reference files inside a skill. Anthropic advises against fetching context through code and injecting it directly into prompts.
Anthropic's findings are consistent with recent research. A study by Vals found that agent teams can be faster than single agents but don't deliver higher quality. Scaling from ten to 30 or 100 agents brought no quality gains but significantly higher costs, according to the study. OpenAI developer Eric Provencher has also questioned the value of agent swarms, calling them "almost certainly" wasted money.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
Subscribe now
Auto-fetched and translated by AI Atlas. All rights belong to the original author.