How we built our multi-agent research system
打开互动全文版（逐段中英对照 + 图/公式 + 论文问答）→我们的研究功能使用多个 Claude 智能体更有效地探索复杂主题。我们分享了构建该系统时遇到的工程挑战和学到的经验。Claude 现在具备研究能力，可以搜索网络、Google Workspace 以及任何集成，以完成复杂任务。这个多智能体系统从原型到生产的历程，让我们在系统架构、工具设计和提示工程方面学到了关键经验。多智能体系统由多个智能体（LLM 在循环中自主使用工具）协同工作组成。我们的研究功能涉及一个智能体，它根据用户查询规划研究过程，然后使用工具创建并行搜索信息的智能体。具有多个智能体的系统在智能体协调、评估和可靠性方面带来了新的挑战。
Our Research feature uses multiple Claude agents to explore complex topics more effectively. We share the engineering challenges and the lessons we learned from building this system. Claude now has Research capabilities that allow it to search across the web, Google Workspace, and any integrations to accomplish complex tasks. The journey of this multi-agent system from prototype to production taught us critical lessons about system architecture, tool design, and prompt engineering. A multi-agent system consists of multiple agents (LLMs autonomously using tools in a loop) working together. Our Research feature involves an agent that plans a research process based on user queries, and then uses tools to create parallel agents that search for information simultaneously. Systems with multiple agents introduce new challenges in agent coordination, evaluation, and reliability.
我们的研究功能使用多个 Claude 智能体更有效地探索复杂主题。我们分享了构建此系统时遇到的工程挑战以及从中汲取的经验教训。
Our Research feature uses multiple Claude agents to explore complex topics more effectively. We share the engineering challenges and the lessons we learned from building this system.
Claude 现在具备研究能力，使其能够搜索网络、Google Workspace 以及任何集成，以完成复杂任务。
Claude now has Research capabilities that allow it to search across the web, Google Workspace, and any integrations to accomplish complex tasks.
这个多智能体系统从原型到生产的过程，教会了我们关于系统架构、工具设计和提示工程的关键经验。多智能体系统由多个智能体（在循环中自主使用工具的 LLM）协同工作组成。我们的研究功能涉及一个智能体，它根据用户查询规划研究过程，然后使用工具创建并行智能体同时搜索信息。多智能体系统引入了智能体协调、评估和可靠性方面的新挑战。
The journey of this multi-agent system from prototype to production taught us critical lessons about system architecture, tool design, and prompt engineering. A multi-agent system consists of multiple agents (LLMs autonomously using tools in a loop) working together. Our Research feature involves an agent that plans a research process based on user queries, and then uses tools to create parallel agents that search for information simultaneously. Systems with multiple agents introduce new challenges in agent coordination, evaluation, and reliability.
本文分解了对我们有效的原则——希望你在构建自己的多智能体系统时发现它们有用。
This post breaks down the principles that worked for us—we hope you'll find them useful to apply when building your own multi-agent systems.
研究工作涉及开放性问题，很难提前预测所需的步骤。你无法为探索复杂主题硬编码固定路径，因为这一过程本质上是动态且路径依赖的。当人们进行研究时，他们往往会根据发现不断调整方法，追踪调查过程中涌现的线索。
Research work involves open-ended problems where it’s very difficult to predict the required steps in advance. You can’t hardcode a fixed path for exploring complex topics, as the process is inherently dynamic and path-dependent. When people conduct research, they tend to continuously update their approach based on discoveries, following leads that emerge during investigation.
这种不可预测性使得 AI 智能体特别适合研究任务。研究需要灵活性，以便在调查展开时转向或探索横向联系。模型必须自主运行多个回合，根据中间结果决定追求哪些方向。线性的单次流水线无法处理这些任务。
This unpredictability makes AI agents particularly well-suited for research tasks. Research demands the flexibility to pivot or explore tangential connections as the investigation unfolds. The model must operate autonomously for many turns, making decisions about which directions to pursue based on intermediate findings. A linear, one-shot pipeline cannot handle these tasks.
搜索的本质是压缩：从海量信息中提炼洞见。子智能体通过并行运行并拥有各自的上下文窗口来促进压缩，同时探索问题的不同方面，然后为首席研究智能体浓缩最重要的词元。每个子智能体还提供了关注点分离——不同的工具、提示和探索轨迹——这减少了路径依赖性，并实现了彻底、独立的调查。
The essence of search is compression: distilling insights from a vast corpus. Subagents facilitate compression by operating in parallel with their own context windows, exploring different aspects of the question simultaneously before condensing the most important tokens for the lead research agent. Each subagent also provides separation of concerns—distinct tools, prompts, and exploration trajectories—which reduces path dependency and enables thorough, independent investigations.
一旦智能达到某个阈值，多智能体系统就成为扩展性能的关键方式。例如，尽管在过去 10 万年中个体人类变得更聪明，但在信息时代，人类社会的能力却呈指数级增长，这是因为我们的集体智慧和协调能力。即使是通用智能体在作为个体运行时也面临限制；智能体群体可以完成更多任务。
Once intelligence reaches a threshold, multi-agent systems become a vital way to scale performance. For instance, although individual humans have become more intelligent in the last 100,000 years, human societies have become exponentially more capable in the information age because of our collective intelligence and ability to coordinate. Even generally-intelligent agents face limits when operating as individuals; groups of agents can accomplish far more.
我们的内部评估显示，多智能体研究系统尤其擅长广度优先的查询，这些查询需要同时追求多个独立方向。我们发现，以 Claude Opus 4 作为首席智能体、Claude Sonnet 4 作为子智能体的多智能体系统，在我们的内部研究评估中比单智能体 Claude Opus 4 性能高出 90.2%。例如，当被要求找出信息科技标普 500 指数中所有公司的董事会成员时，多智能体系统通过将任务分解给子智能体找到了正确答案，而单智能体系统由于缓慢的顺序搜索未能找到答案。
Our internal evaluations show that multi-agent research systems excel especially for breadth-first queries that involve pursuing multiple independent directions simultaneously. We found that a multi-agent system with Claude Opus 4 as the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on our internal research eval. For example, when asked to identify all the board members of the companies in the Information Technology S&P 500, the multi-agent system found the correct answers by decomposing this into tasks for subagents, while the single agent system failed to find the answer with slow, sequential searches.
多智能体系统之所以有效，主要是因为它们能够花费足够的词元来解决问题。在我们的分析中，三个因素解释了 BrowseComp 评估（测试浏览智能体定位难以找到信息的能力）中 95%的性能差异。我们发现，词元使用量本身解释了 80%的差异，工具调用次数和模型选择是另外两个解释因素。这一发现验证了我们的架构，该架构将工作分布在具有独立上下文窗口的智能体之间，以增加并行推理的能力。最新的 Claude 模型充当了词元使用的大效率倍增器，因为升级到 Claude Sonnet 4 带来的性能提升比将 Claude Sonnet 3.7 的词元预算翻倍还要大。多智能体架构有效地扩展了词元使用，以应对超出单智能体限制的任务。
Multi-agent systems work mainly because they help spend enough tokens to solve the problem. In our analysis, three factors explained 95% of the performance variance in the BrowseComp evaluation (which tests the ability of browsing agents to locate hard-to-find information). We found that token usage by itself explains 80% of the variance, with the number of tool calls and the model choice as the two other explanatory factors. This finding validates our architecture that distributes work across agents with separate context windows to add more capacity for parallel reasoning. The latest Claude models act as large efficiency multipliers on token use, as upgrading to Claude Sonnet 4 is a larger performance gain than doubling the token budget on Claude Sonnet 3.7. Multi-agent architectures effectively scale token usage for tasks that exceed the limits of single agents.
但有一个缺点：在实践中，这些架构会快速消耗词元。在我们的数据中，智能体通常使用约 4 倍于聊天交互的词元，而多智能体系统使用约 15 倍于聊天的词元。为了经济可行性，多智能体系统需要任务的价值足够高，以支付性能提升的成本。此外，某些要求所有智能体共享相同上下文或涉及智能体之间大量依赖关系的领域，目前并不适合多智能体系统。例如，大多数编码任务的可并行化任务比研究少，且 LLM 智能体在实时协调和委托给其他智能体方面尚不擅长。我们发现，多智能体系统擅长那些涉及大量并行化、信息超出单个上下文窗口以及需要与众多复杂工具交互的高价值任务。
There is a downside: in practice, these architectures burn through tokens fast. In our data, agents typically use about 4× more tokens than chat interactions, and multi-agent systems use about 15× more tokens than chats. For economic viability, multi-agent systems require tasks where the value of the task is high enough to pay for the increased performance. Further, some domains that require all agents to share the same context or involve many dependencies between agents are not a good fit for multi-agent systems today. For instance, most coding tasks involve fewer truly parallelizable tasks than research, and LLM agents are not yet great at coordinating and delegating to other agents in real time. We’ve found that multi-agent systems excel at valuable tasks that involve heavy parallelization, information that exceeds single context windows, and interfacing with numerous complex tools.
我们的研究系统采用多智能体架构，遵循编排器-工作者模式，由一个主导智能体协调流程，同时将任务委派给并行运行的专业子智能体。
Our Research system uses a multi-agent architecture with an orchestrator-worker pattern, where a lead agent coordinates the process while delegating to specialized subagents that operate in parallel.
多智能体架构的实际运作：用户查询流经主导智能体，该智能体创建专业子智能体，以并行搜索不同方面。
The multi-agent architecture in action: user queries flow through a lead agent that creates specialized subagents to search for different aspects in parallel.
当用户提交查询时，主导智能体分析查询、制定策略，并生成子智能体同时探索不同方面。如上图所示，子智能体通过迭代使用搜索工具收集信息（本例中为 2025 年 AI 智能体公司），充当智能过滤器，然后将公司列表返回给主导智能体，以便其编译最终答案。
When a user submits a query, the lead agent analyzes it, develops a strategy, and spawns subagents to explore different aspects simultaneously. As shown in the diagram above, the subagents act as intelligent filters by iteratively using search tools to gather information, in this case on AI agent companies in 2025, and then returning a list of companies to the lead agent so it can compile a final answer.
传统方法使用检索增强生成（RAG）进行静态检索，即获取与输入查询最相似的一些文本块，并用这些文本块生成响应。相比之下，我们的架构采用多步搜索，动态查找相关信息，适应新发现，并分析结果以制定高质量答案。
Traditional approaches using Retrieval Augmented Generation (RAG) use static retrieval. That is, they fetch some set of chunks that are most similar to an input query and use these chunks to generate a response. In contrast, our architecture uses a multi-step search that dynamically finds relevant information, adapts to new findings, and analyzes results to formulate high-quality answers.
过程图展示了多智能体研究系统的完整工作流程。当用户提交查询时，系统创建一个 LeadResearcher 智能体，进入迭代研究过程。LeadResearcher 首先思考方法，并将其计划保存到 Memory 中以持久化上下文，因为如果上下文窗口超过 200,000 个 token，计划将被截断，保留计划至关重要。然后，它创建具有特定研究任务的专业子智能体（图中显示两个，但可以是任意数量）。每个子智能体独立执行网络搜索，使用交错思考评估工具结果，并将发现返回给 LeadResearcher。LeadResearcher 综合这些结果，并决定是否需要更多研究——如果需要，它可以创建额外的子智能体或优化其策略。一旦收集到足够信息，系统退出研究循环，并将所有发现传递给 CitationAgent，该智能体处理文档和研究报告，以确定引用的具体位置。这确保所有声明都正确归因于其来源。最终的研究结果（附有引用）随后返回给用户。
Process diagram showing the complete workflow of our multi-agent Research system. When a user submits a query, the system creates a LeadResearcher agent that enters an iterative research process. The LeadResearcher begins by thinking through the approach and saving its plan to Memory to persist the context, since if the context window exceeds 200,000 tokens it will be truncated and it is important to retain the plan. It then creates specialized Subagents (two are shown here, but it can be any number) with specific research tasks. Each Subagent independently performs web searches, evaluates tool results using interleaved thinking, and returns findings to the LeadResearcher. The LeadResearcher synthesizes these results and decides whether more research is needed—if so, it can create additional subagents or refine its strategy. Once sufficient information is gathered, the system exits the research loop and passes all findings to a CitationAgent, which processes the documents and research report to identify specific locations for citations. This ensures all claims are properly attributed to their sources. The final research results, complete with citations, are then returned to the user.
多智能体系统与单智能体系统存在关键差异，包括协调复杂性的快速增长。早期智能体会犯诸如为简单查询生成 50 个子智能体、无休止地搜索不存在的来源、以及用过多更新互相干扰等错误。由于每个智能体由提示词驱动，提示工程是我们改进这些行为的主要手段。以下是我们总结的智能体提示原则：
Multi-agent systems have key differences from single-agent systems, including a rapid growth in coordination complexity. Early agents made errors like spawning 50 subagents for simple queries, scouring the web endlessly for nonexistent sources, and distracting each other with excessive updates. Since each agent is steered by a prompt, prompt engineering was our primary lever for improving these behaviors. Below are some principles we learned for prompting agents:
1. 像你的智能体一样思考。要迭代提示词，你必须理解其效果。为此，我们使用系统中完全相同的提示词和工具在 Console 中构建模拟，然后逐步观察智能体工作。这立即揭示了失败模式：智能体在已有足够结果时继续运行、使用过于冗长的搜索查询、或选择错误的工具。有效的提示依赖于对智能体建立准确的思维模型，这能使最具影响力的改变变得显而易见。
1. Think like your agents.To iterate on prompts, you must understand their effects. To help us do this, we built simulations using our Console with the exact prompts and tools from our system, then watched agents work step-by-step. This immediately revealed failure modes: agents continuing when they already had sufficient results, using overly verbose search queries, or selecting incorrect tools. Effective prompting relies on developing an accurate mental model of the agent, which can make the most impactful changes obvious.
2. 教导协调者如何委派任务。在我们的系统中，主导智能体将查询分解为子任务并描述给子智能体。每个子智能体需要目标、输出格式、关于使用哪些工具和来源的指导，以及明确的任务边界。没有详细的任务描述，智能体会重复工作、留下空白或无法找到必要信息。我们最初允许主导智能体给出简单简短的指令，如“研究半导体短缺”，但发现这些指令往往过于模糊，导致子智能体误解任务或执行与其他智能体完全相同的搜索。例如，一个子智能体探索了 2021 年汽车芯片危机，而另外两个子智能体重复了调查当前 2025 年供应链的工作，没有有效的分工。
2. Teach the orchestrator how to delegate. In our system, the lead agent decomposes queries into subtasks and describes them to subagents. Each subagent needs an objective, an output format, guidance on the tools and sources to use, and clear task boundaries. Without detailed task descriptions, agents duplicate work, leave gaps, or fail to find necessary information. We started by allowing the lead agent to give simple, short instructions like 'research the semiconductor shortage,' but found these instructions often were vague enough that subagents misinterpreted the task or performed the exact same searches as other agents. For instance, one subagent explored the 2021 automotive chip crisis while 2 others duplicated work investigating current 2025 supply chains, without an effective division of labor.
3. 根据查询复杂性调整投入。智能体难以判断不同任务的适当投入，因此我们在提示中嵌入了规模规则。简单的事实查找只需 1 个智能体进行 3-10 次工具调用，直接比较可能需要 2-4 个子智能体各调用 10-15 次，而复杂研究可能使用超过 10 个具有明确分工的子智能体。这些明确的指导方针帮助主导智能体高效分配资源，并防止对简单查询过度投入，这是我们早期版本中常见的失败模式。
3. Scale effort to query complexity.Agents struggle to judge appropriate effort for different tasks, so we embedded scaling rules in the prompts. Simple fact-finding requires just 1 agent with 3-10 tool calls, direct comparisons might need 2-4 subagents with 10-15 calls each, and complex research might use more than 10 subagents with clearly divided responsibilities. These explicit guidelines help the lead agent allocate resources efficiently and prevent overinvestment in simple queries, which was a common failure mode in our early versions.
4. 工具设计和选择至关重要。智能体-工具接口与人类-计算机接口同等重要。使用正确的工具是高效的——通常也是严格必要的。例如，一个智能体在网络上搜索仅存在于 Slack 中的上下文，从一开始就注定失败。当 MCP 服务器让模型访问外部工具时，这个问题会加剧，因为智能体会遇到质量参差不齐的工具描述。我们给智能体明确的启发式规则：例如，首先检查所有可用工具，将工具使用与用户意图匹配，搜索网络进行广泛的外部探索，或优先选择专用工具而非通用工具。糟糕的工具描述可能使智能体完全走错方向，因此每个工具都需要明确的目的和清晰的描述。
4. Tool design and selection are critical.Agent-tool interfaces are as critical as human-computer interfaces. Using the right tool is efficient—often, it’s strictly necessary. For instance, an agent searching the web for context that only exists in Slack is doomed from the start. With MCP servers that give the model access to external tools, this problem compounds, as agents encounter unseen tools with descriptions of wildly varying quality. We gave our agents explicit heuristics: for example, examine all available tools first, match tool usage to user intent, search the web for broad external exploration, or prefer specialized tools over generic ones. Bad tool descriptions can send agents down completely wrong paths, so each tool needs a distinct purpose and a clear description.
5. 让智能体自我改进。我们发现 Claude 4 模型可以成为优秀的提示工程师。当给定提示词和失败模式时，它们能够诊断智能体失败的原因并提出改进建议。我们甚至创建了一个工具测试智能体——当给定一个有缺陷的 MCP 工具时，它会尝试使用该工具，然后重写工具描述以避免失败。通过数十次测试该工具，这个智能体发现了关键细微差别和错误。这一改进工具可用性的过程使后续使用新描述的智能体任务完成时间减少了 40%，因为它们能够避免大多数错误。
5. Let agents improve themselves. We found that the Claude 4 models can be excellent prompt engineers. When given a prompt and a failure mode, they are able to diagnose why the agent is failing and suggest improvements. We even created a tool-testing agent—when given a flawed MCP tool, it attempts to use the tool and then rewrites the tool description to avoid failures. By testing the tool dozens of times, this agent found key nuances and bugs. This process for improving tool ergonomics resulted in a 40% decrease in task completion time for future agents using the new description, because they were able to avoid most mistakes.
6. 先宽泛，再聚焦。搜索策略应模仿人类专家研究：在深入细节之前先探索全局。智能体往往默认使用过长、过于具体的查询，返回结果很少。我们通过提示智能体从简短、宽泛的查询开始，评估可用信息，然后逐步缩小焦点来对抗这种倾向。
6. Start wide, then narrow down. Search strategy should mirror expert human research: explore the landscape before drilling into specifics. Agents often default to overly long, specific queries that return few results. We counteracted this tendency by prompting agents to start with short, broad queries, evaluate what’s available, then progressively narrow focus.
7. 引导思考过程。扩展思考模式使 Claude 在可见的思考过程中输出额外 token，可作为可控的草稿板。主导智能体使用思考来规划方法，评估哪些工具适合任务，确定查询复杂性和子智能体数量，并定义每个子智能体的角色。我们的测试表明，扩展思考改进了指令遵循、推理和效率。子智能体也会进行规划，然后在工具结果后使用交错思考来评估质量、识别差距并优化下一个查询。这使得子智能体在适应任何任务时更加有效。
7. Guide the thinking process.Extended thinking mode, which leads Claude to output additional tokens in a visible thinking process, can serve as a controllable scratchpad. The lead agent uses thinking to plan its approach, assessing which tools fit the task, determining query complexity and subagent count, and defining each subagent’s role. Our testing showed that extended thinking improved instruction-following, reasoning, and efficiency. Subagents also plan, then use interleaved thinking after tool results to evaluate quality, identify gaps, and refine their next query. This makes subagents more effective in adapting to any task.
8. 并行工具调用改变速度和性能。复杂的研究任务自然涉及探索多个来源。我们的早期智能体执行顺序搜索，速度极慢。为了速度，我们引入了两种并行化：（1）主导智能体并行启动 3-5 个子智能体，而非串行；（2）子智能体并行使用 3 个以上工具。这些改变将复杂查询的研究时间减少了高达 90%，使研究能在几分钟内完成原本需要数小时的工作，同时覆盖比其他系统更多的信息。
8. Parallel tool calling transforms speed and performance. Complex research tasks naturally involve exploring many sources. Our early agents executed sequential searches, which was painfully slow. For speed, we introduced two kinds of parallelization: (1) the lead agent spins up 3-5 subagents in parallel rather than serially; (2) the subagents use 3+ tools in parallel. These changes cut research time by up to 90% for complex queries, allowing Research to do more work in minutes instead of hours while covering more information than other systems.
我们的提示策略侧重于灌输良好的启发式规则而非僵化的规则。我们研究了熟练人类如何进行研究任务，并将这些策略编码到提示中——例如将困难问题分解为更小的任务、仔细评估来源质量、根据新信息调整搜索方法、以及认识到何时专注于深度（详细调查一个主题）与广度（并行探索多个主题）。我们还通过设置明确的护栏来主动减轻意外副作用，防止智能体失控。最后，我们专注于具有可观测性和测试用例的快速迭代循环。
Our prompting strategy focuses on instilling good heuristics rather than rigid rules. We studied how skilled humans approach research tasks and encoded these strategies in our prompts—strategies like decomposing difficult questions into smaller tasks, carefully evaluating the quality of sources, adjusting search approaches based on new information, and recognizing when to focus on depth (investigating one topic in detail) vs. breadth (exploring many topics in parallel). We also proactively mitigated unintended side effects by setting explicit guardrails to prevent the agents from spiraling out of control. Finally, we focused on a fast iteration loop with observability and test cases.
良好的评估对于构建可靠的 AI 应用至关重要，智能体也不例外。然而，评估多智能体系统带来了独特的挑战。传统评估通常假设 AI 每次都遵循相同的步骤：给定输入 X，系统应遵循路径 Y 以产生输出 Z。但多智能体系统并非如此。即使起点相同，智能体也可能采取完全不同的有效路径来达成目标。一个智能体可能搜索三个来源，而另一个搜索十个，或者它们可能使用不同的工具来找到相同的答案。由于我们并不总是知道正确的步骤是什么，通常无法仅检查智能体是否遵循了我们预先规定的“正确”步骤。相反，我们需要灵活的评估方法，既能判断智能体是否达成了正确的结果，同时也能评估其过程是否合理。
Good evaluations are essential for building reliable AI applications, and agents are no different. However, evaluating multi-agent systems presents unique challenges. Traditional evaluations often assume that the AI follows the same steps each time: given input X, the system should follow path Y to produce output Z. But multi-agent systems don't work this way. Even with identical starting points, agents might take completely different valid paths to reach their goal. One agent might search three sources while another searches ten, or they might use different tools to find the same answer. Because we don’t always know what the right steps are, we usually can't just check if agents followed the “correct” steps we prescribed in advance. Instead, we need flexible evaluation methods that judge whether agents achieved the right outcomes while also following a reasonable process.
立即从小样本开始评估。在智能体开发的早期阶段，由于存在大量唾手可得的改进点，变化往往会产生显著影响。一个提示词的调整可能将成功率从 30%提升到 80%。在效果如此显著的情况下，只需几个测试用例就能发现变化。我们从一组约 20 个代表实际使用模式的查询开始。测试这些查询通常能让我们清晰地看到变化的影响。我们经常听到 AI 开发团队推迟创建评估，因为他们认为只有包含数百个测试用例的大型评估才有用。然而，最好立即从小规模测试开始，使用少量示例，而不是等到能够构建更全面的评估时才进行。
Start evaluating immediately with small samples. In early agent development, changes tend to have dramatic impacts because there is abundant low-hanging fruit. A prompt tweak might boost success rates from 30% to 80%. With effect sizes this large, you can spot changes with just a few test cases. We started with a set of about 20 queries representing real usage patterns. Testing these queries often allowed us to clearly see the impact of changes. We often hear that AI developer teams delay creating evals because they believe that only large evals with hundreds of test cases are useful. However, it’s best to start with small-scale testing right away with a few examples, rather than delaying until you can build more thorough evals.
LLM 作为评判者的评估在实施得当的情况下具有可扩展性。研究型输出难以通过编程方式评估，因为它们通常是自由形式的文本，很少有一个唯一正确的答案。LLM 自然适合对输出进行评分。我们使用了一个 LLM 评判者，根据评分标准中的各项标准对每个输出进行评估：事实准确性（声明是否与来源匹配？）、引用准确性（引用的来源是否与声明匹配？）、完整性（是否涵盖了所有要求的方面？）、来源质量（是否优先使用主要来源而非低质量的次要来源？）以及工具效率（是否以合理的次数使用了正确的工具？）。我们尝试了多个评判者来评估每个组成部分，但发现使用单个 LLM 调用、单个提示词输出 0.0-1.0 的分数和通过/不及格等级是最一致且与人类判断最吻合的。当评估测试用例确实有明确答案时，这种方法尤其有效，我们可以让 LLM 评判者简单地检查答案是否正确（例如，它是否准确列出了研发预算排名前三的制药公司？）。使用 LLM 作为评判者使我们能够可扩展地评估数百个输出。
LLM-as-judge evaluation scales when done well. Research outputs are difficult to evaluate programmatically, since they are free-form text and rarely have a single correct answer. LLMs are a natural fit for grading outputs. We used an LLM judge that evaluated each output against criteria in a rubric: factual accuracy (do claims match sources?), citation accuracy (do the cited sources match the claims?), completeness (are all requested aspects covered?), source quality (did it use primary sources over lower-quality secondary sources?), and tool efficiency (did it use the right tools a reasonable number of times?). We experimented with multiple judges to evaluate each component, but found that a single LLM call with a single prompt outputting scores from 0.0-1.0 and a pass-fail grade was the most consistent and aligned with human judgements. This method was especially effective when the eval test cases did have a clear answer, and we could use the LLM judge to simply check if the answer was correct (i.e. did it accurately list the pharma companies with the top 3 largest R&D budgets?). Using an LLM as a judge allowed us to scalably evaluate hundreds of outputs.
人工评估能捕捉自动化遗漏的问题。测试智能体的人员会发现评估遗漏的边缘情况。这些包括对不常见查询的幻觉回答、系统故障或微妙的来源选择偏差。在我们的案例中，人工测试人员注意到我们的早期智能体一致地选择 SEO 优化的内容农场，而不是权威但排名较低的资源，如学术 PDF 或个人博客。在提示词中添加来源质量启发式规则有助于解决这个问题。即使在自动化评估的世界中，手动测试仍然至关重要。
Human evaluation catches what automation misses. People testing agents find edge cases that evals miss. These include hallucinated answers on unusual queries, system failures, or subtle source selection biases. In our case, human testers noticed that our early agents consistently chose SEO-optimized content farms over authoritative but less highly-ranked sources like academic PDFs or personal blogs. Adding source quality heuristics to our prompts helped resolve this issue. Even in a world of automated evaluations, manual testing remains essential.
多智能体系统具有涌现行为，这些行为无需特定编程即可出现。例如，对主导智能体的微小改变可能不可预测地改变子智能体的行为。成功需要理解交互模式，而不仅仅是单个智能体的行为。因此，这些智能体的最佳提示词不仅仅是严格的指令，而是定义分工、问题解决方法和努力预算的协作框架。要正确实现这一点，依赖于仔细的提示词设计和工具设计、可靠的启发式规则、可观测性以及紧密的反馈循环。请参阅我们的 Cookbook 中的开源提示词，以获取我们系统中的示例提示词。
Multi-agent systems have emergent behaviors, which arise without specific programming. For instance, small changes to the lead agent can unpredictably change how subagents behave. Success requires understanding interaction patterns, not just individual agent behavior. Therefore, the best prompts for these agents are not just strict instructions, but frameworks for collaboration that define the division of labor, problem-solving approaches, and effort budgets. Getting this right relies on careful prompting and tool design, solid heuristics, observability, and tight feedback loops.See the open-source prompts in our Cookbook for example prompts from our system.
在传统软件中，一个错误可能破坏某个功能、降低性能或导致服务中断。而在智能体式系统中，微小的变化会级联成巨大的行为变化，这使得为需要在长时间运行过程中维护状态的复杂智能体编写代码变得异常困难。
In traditional software, a bug might break a feature, degrade performance, or cause outages. In agentic systems, minor changes cascade into large behavioral changes, which makes it remarkably difficult to write code for complex agents that must maintain state in a long-running process.
智能体是有状态的，且错误会累积。智能体可以长时间运行，在多次工具调用中维护状态。这意味着我们需要持久地执行代码并沿途处理错误。如果没有有效的缓解措施，微小的系统故障对智能体来说可能是灾难性的。当错误发生时，我们不能简单地从头开始重启：重启成本高昂且让用户感到沮丧。相反，我们构建了能够从智能体出错时的位置恢复的系统。我们还利用模型的智能来优雅地处理问题：例如，让智能体知道工具何时失败并让其适应，效果出奇地好。我们将基于 Claude 构建的 AI 智能体的适应性与确定性防护措施（如重试逻辑和定期检查点）相结合。
Agents are stateful and errors compound.Agents can run for long periods of time, maintaining state across many tool calls. This means we need to durably execute code and handle errors along the way. Without effective mitigations, minor system failures can be catastrophic for agents. When errors occur, we can't just restart from the beginning: restarts are expensive and frustrating for users. Instead, we built systems that can resume from where the agent was when the errors occurred. We also use the model’s intelligence to handle issues gracefully: for instance, letting the agent know when a tool is failing and letting it adapt works surprisingly well. We combine the adaptability of AI agents built on Claude with deterministic safeguards like retry logic and regular checkpoints.
调试需要新的方法。智能体做出动态决策，并且在不同的运行之间是非确定性的，即使使用相同的提示也是如此。这使得调试更加困难。例如，用户报告智能体“找不到明显的信息”，但我们无法看到原因。是智能体使用了糟糕的搜索查询？选择了不良来源？还是遇到了工具故障？添加完整的生产追踪让我们能够诊断智能体失败的原因并系统地修复问题。除了标准的可观测性之外，我们还监控智能体的决策模式和交互结构——所有这些都不监控单个对话的内容，以维护用户隐私。这种高层级的可观测性帮助我们诊断根本原因、发现意外行为并修复常见故障。
Debugging benefits from new approaches.Agents make dynamic decisions and are non-deterministic between runs, even with identical prompts. This makes debugging harder. For instance, users would report agents “not finding obvious information,” but we couldn't see why. Were the agents using bad search queries? Choosing poor sources? Hitting tool failures? Adding full production tracing let us diagnose why agents failed and fix issues systematically. Beyond standard observability, we monitor agent decision patterns and interaction structures—all without monitoring the contents of individual conversations, to maintain user privacy. This high-level observability helped us diagnose root causes, discover unexpected behaviors, and fix common failures.
部署需要仔细协调。智能体系统是高度有状态的提示、工具和执行逻辑的网络，几乎持续运行。这意味着每当我们部署更新时，智能体可能处于其流程中的任何位置。因此，我们需要防止我们善意的代码更改破坏现有的智能体。我们不能同时将所有智能体更新到新版本。相反，我们使用彩虹部署来避免中断正在运行的智能体，通过逐步将流量从旧版本转移到新版本，同时保持两者同时运行。
Deployment needs careful coordination. Agent systems are highly stateful webs of prompts, tools, and execution logic that run almost continuously. This means that whenever we deploy updates, agents might be anywhere in their process. We therefore need to prevent our well-meaning code changes from breaking existing agents. We can’t update every agent to the new version at the same time. Instead, we use rainbow deployments to avoid disrupting running agents, by gradually shifting traffic from old to new versions while keeping both running simultaneously.
同步执行造成瓶颈。目前，我们的主导智能体同步执行子智能体，等待每组子智能体完成后再继续。这简化了协调，但在智能体之间的信息流中造成了瓶颈。例如，主导智能体无法引导子智能体，子智能体无法相互协调，整个系统可能因等待单个子智能体完成搜索而被阻塞。异步执行将实现额外的并行性：智能体并发工作，并在需要时创建新的子智能体。但这种异步性在结果协调、状态一致性和子智能体之间的错误传播方面增加了挑战。随着模型能够处理更长、更复杂的研究任务，我们预计性能提升将证明这种复杂性的合理性。
Synchronous execution creates bottlenecks. Currently, our lead agents execute subagents synchronously, waiting for each set of subagents to complete before proceeding. This simplifies coordination, but creates bottlenecks in the information flow between agents. For instance, the lead agent can’t steer subagents, subagents can’t coordinate, and the entire system can be blocked while waiting for a single subagent to finish searching. Asynchronous execution would enable additional parallelism: agents working concurrently and creating new subagents when needed. But this asynchronicity adds challenges in result coordination, state consistency, and error propagation across the subagents. As models can handle longer and more complex research tasks, we expect the performance gains will justify the complexity.
在构建 AI 智能体时，最后一英里往往占据了大部分旅程。在开发者机器上能运行的代码库需要大量的工程工作才能成为可靠的生产系统。智能体系统中错误的复合性质意味着，对于传统软件来说的小问题可能会完全破坏智能体。一个步骤的失败可能导致智能体探索完全不同的轨迹，从而产生不可预测的结果。由于本文中描述的所有原因，原型与生产之间的差距往往比预期的要大。
When building AI agents, the last mile often becomes most of the journey. Codebases that work on developer machines require significant engineering to become reliable production systems. The compound nature of errors in agentic systems means that minor issues for traditional software can derail agents entirely. One step failing can cause agents to explore entirely different trajectories, leading to unpredictable outcomes. For all the reasons described in this post, the gap between prototype and production is often wider than anticipated.
尽管存在这些挑战，多智能体系统已被证明对开放式研究任务有价值。用户表示，Claude 帮助他们找到了未曾考虑过的商业机会，导航复杂的医疗保健选项，解决棘手的技术错误，并通过发现他们独自无法找到的研究联系，节省了多达数天的工作。通过精心的工程、全面的测试、注重细节的提示和工具设计、稳健的操作实践，以及研究、产品和工程团队之间对当前智能体能力的深刻理解与紧密协作，多智能体研究系统可以大规模可靠地运行。我们已经看到这些系统正在改变人们解决复杂问题的方式。
Despite these challenges, multi-agent systems have proven valuable for open-ended research tasks. Users have said that Claude helped them find business opportunities they hadn’t considered, navigate complex healthcare options, resolve thorny technical bugs, and save up to days of work by uncovering research connections they wouldn't have found alone. Multi-agent research systems can operate reliably at scale with careful engineering, comprehensive testing, detail-oriented prompt and tool design, robust operational practices, and tight collaboration between research, product, and engineering teams who have a strong understanding of current agent capabilities. We're already seeing these systems transform how people solve complex problems.
一个 Clio 嵌入图显示了人们今天使用 Research 功能的最常见方式。前五大使用类别是：跨专业领域开发软件系统（10%），开发和优化专业及技术内容（8%），制定业务增长和收入生成策略（8%），协助学术研究和教育材料开发（7%），以及研究和验证关于人物、地点或组织的信息（5%）。
A Clio embedding plot showing the most common ways people are using the Research feature today. The top use case categories are developing software systems across specialized domains (10%), develop and optimize professional and technical content (8%), develop business growth and revenue generation strategies (8%), assist with academic research and educational material development (7%), and research and verify information about people, places, or organizations (5%).
本文由 Jeremy Hadfield、Barry Zhang、Kenneth Lien、Florian Scholz、Jeremy Fox 和 Daniel Ford 撰写。这项工作反映了 Anthropic 多个团队的集体努力，使研究功能成为可能。特别感谢 Anthropic 应用工程团队，他们的奉献精神使这个复杂的多智能体系统得以投入生产。我们也感谢早期用户提供的宝贵反馈。
Written by Jeremy Hadfield, Barry Zhang, Kenneth Lien, Florian Scholz, Jeremy Fox, and Daniel Ford. This work reflects the collective efforts of several teams across Anthropic who made the Research feature possible. Special thanks go to the Anthropic apps engineering team, whose dedication brought this complex multi-agent system to production. We're also grateful to our early users for their excellent feedback.