{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-20T12:00:59.032Z","headline":"Anthropic 发布 Claude 电商智能体构建指南及参考实现","description":"Anthropic 发布构建电商智能体的指南，总结架构、延迟与成本优化、生产运维三部分实践，涉及购物和商家两类智能体，企业客户部署后出现购物车变大和卖家运营效率提升。核心建议包括用单个智能体加技能而非子智能体、将 UI 组件做成工具、用提示词缓存实现 90-99% 命中率、在 harness 中强制安全规则，并开源参考实现 anthropics/commerce-agents。","url":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","mainEntityOfPage":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","datePublished":"2026-09-01T16:00:00.000Z","dateModified":"2026-09-01T16:00:00.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://claude.com/blog/the-anatomy-of-effective-commerce-agents","https://aihot.virxact.com/items/cmtkcffah018zrog0zokk6s30"],"canonicalUrl":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","directAnswer":{"@type":"Answer","text":"Anthropic 发布面向电商智能体开发者的指南及参考实现，内容覆盖架构、延迟与成本、内存、安全、评估和组织扩展，并讨论购物端与商家端两类应用。","url":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","dateCreated":"2026-09-01T16:00:00.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"Claude：Blog（网页） source article","url":"https://claude.com/blog/the-anatomy-of-effective-commerce-agents","datePublished":"2026-09-01T16:00:00.000Z","provider":{"@type":"Organization","name":"Claude：Blog（网页）","url":"https://claude.com/blog/the-anatomy-of-effective-commerce-agents"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmtkcffah018zrog0zokk6s30","datePublished":"2026-09-01T16:00:00.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmtkcffah018zrog0zokk6s30"}}],"aggregationSource":"Claude：Blog（网页）","originalPublisher":{"name":"Claude：Blog（网页）","url":"https://claude.com/blog/the-anatomy-of-effective-commerce-agents"},"geoDeepAnswer":null,"article":{"id":"cmtkcffah018zrog0zokk6s30","slug":"cmtkcffah018zrog0zokk6s30","url":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","title":"Anthropic 发布 Claude 电商智能体构建指南及参考实现","title_en":"","summary":"Anthropic 发布构建电商智能体的指南，总结架构、延迟与成本优化、生产运维三部分实践，涉及购物和商家两类智能体，企业客户部署后出现购物车变大和卖家运营效率提升。核心建议包括用单个智能体加技能而非子智能体、将 UI 组件做成工具、用提示词缓存实现 90-99% 命中率、在 harness 中强制安全规则，并开源参考实现 anthropics/commerce-agents。","source":"Claude：Blog（网页）","sourceUrl":"https://claude.com/blog/the-anatomy-of-effective-commerce-agents","aiHotUrl":"https://aihot.virxact.com/items/cmtkcffah018zrog0zokk6s30","publishedAt":"2026-09-01T16:00:00.000Z","category":"行业动态","score":72,"selected":true,"articleBody":["The architecture, latency & cost techniques, and eval practices for agents that make it easier to buy and sell online.","Over the past year, we've worked with teams across the commerce industry — retailers, marketplaces, travel, entertainment, and telecom providers — to build commerce agents using Claude.","These agents are in production, and enterprise customers have seen larger carts and more efficient seller operations when using them. They also share a simple architecture: Claude in an agent loop equipped with a set of skills, tools, and a strong eval suite.","This post is for the engineers and engineering leaders building these (or other consumer facing) agents. Part 1 covers the architecture, which you decide once. Part 2 covers latency and cost. Part 3 covers production: memory, safety, evals, and scaling the work across an organization.","We define a commerce agent as an agent that simplifies buying and selling across an online catalog.","Some agents face consumers: they search, compare, substitute, and assemble the order. That could be a retail cart, a travel itinerary, a mobile plan change, or seats held for a show. Some agents face the business: they answer questions about sales, run promotions and campaigns, and manage inventory and pricing.","The core architecture is a model in a standard agent loop：https://www.anthropic.com/engineering/building-effective-agents: reasoning about a goal, exploring context, taking actions through tools, learning procedures through skills, asking clarifying questions, and observing the results until the goal is accomplished.","There is no intent router in front of it that segments the conversation and no set of domain specific agents behind it.","A commerce agent has to cover a wide range of capabilities across many categories and intents, which makes it tempting to create one subagent per domain.","In practice this proves suboptimal, because a commerce conversation is one tightly coupled session across multiple intents and turns, and requires considerable shared context.","In a subagent architecture, the orchestrator holds the cart or staged changes, the user's preferences, and the conversation history.","Every handoff to a subagent is a state-lossy operation, which often impacts the quality of the subagent’s response and, consequently, the overall response. On top of that, each handoff can cost several times the tokens and adds seconds of latency.","The domains also rarely separate cleanly. A returns flow might need the order history, the current cart, and the product catalog, meaning a subagent-per-domain approach either duplicates that access everywhere or hands off mid-task.","As models get smarter, they also handle longer context, more skills, and more tools, so the limits behind today's placement rules loosen with each model generation.","Instead, agent skills：https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview give you similar per-domain modularity and context control without the handoff tax, because the skill instructions load into the main agent that already holds the entire history.","In our comparisons across several enterprise deployments, a single agent with skills consistently has outperformed both the one-prompt-for-everything design and the subagent design on quality, and often at a lower cost and latency per task.","Where subagents do earn their place is when the orchestrator can call them as a tool for a narrow or self-contained task that would benefit from its own dedicated context window.","A common production example is a deep-research subagent, where the subagent searches and reads documents, writes and runs code, traverses data models, and hits dead ends. All the work happens inside one or more subagents, and only a compact answer comes back to the orchestrator.","The other exception is a domain that already has its own purpose-built agent. If your pharmacy or financial-services experience runs a dedicated agent with its own compliance surface, the right move can be a hand-off, where that agent takes over the task and works with the user directly through its own loop until the task is done.","The distinction is ownership of the conversation. A hand-off makes the domain agent the user's counterpart, while delegation keeps the orchestrator, bouncing the domain agent in and out within a single turn and degrading on every exchange.","The main factor when deciding whether to put a set of instructions within a system prompt or skill is how often the agent will need it. Loading a skill costs a model turn, so anything the agent needs on most turns generally goes in the system prompt.","This does, however, depend on how your traffic is distributed, and what agent behavior your evals show. A good starting point is that anything relevant to a third or more of your traffic, whether anticipated before launch or observed in production, goes in the system prompt, and the rest goes in skills.","If a skill is predictable from a signal you already have, such as the page the user arrived from, we recommend injecting it from the harness before the first model call and skipping the extra turn to load the skill.","Critical instructions, such as safety and legal rules, brand constraints, and key user facts such as allergies, always go in the system prompt.","For commerce agents, this means product search lives in the prompt, since nearly every session touches it, and skills carry the long tail of features.","In our reference implementation：https://github.com/anthropics/commerce-agents, the shopping agent's prompt holds grounding, cart and checkout semantics, and presentation rules, and the following skills cover the rest: search-discovery, purchase-research, planning-goals, customer-care, and memory-personalization.","The merchant agent splits the same way, with performance-insights, catalog-listings, inventory-operations, pricing-promotions, and marketing-campaigns as its skills, one per operational domain.","Our post on writing effective tools for agents：https://www.anthropic.com/engineering/writing-tools-for-agents covers tool design in general. Two points have mattered most in commerce:","Build agent tools on top of your core systems and logic.","A commerce company already has search and ranking, a cart, a preferences and profile store, an inventory system, promotion and campaign engines, sales analytics, and more, each encoding logic tuned over years and seeing signals the model never will.","The agent's tools should call those systems, not reimplement them, and the tool boundary is where their logic ends and the model's judgment takes over.","For example, when the agent calls search_products , the results should arrive already ranked; its job is to decide which results serve the user's goal, how many to show, and how to present them.","Return the fields the model reasons with and drop the rest. Image URLs on every search row are the usual offender.","As needed, reshape the raw response inside the tool, including appending a next step when it isn't obvious from the data.","This is especially relevant for error scenarios, where the model benefits from instructions instead of error codes. For example, add an error instruction \"Include a product ID when querying availability,\" instead of a generic 403.","Most commerce agent responses are UI components rather than prose, whether a product carousel, an itinerary, a seat map, or a chart. That means the agent has to emit a schema rather than text.","Teams sometimes start by prompting the model to emit custom tags and parsing them on the client-side. This stops working as the surface grows, because:","The pattern that has held up is to make each UI component a tool. The model calls present_products , present_itinerary , or present_plan_comparison with typed arguments; your server validates and enriches the call and emits an event; and your client renders it.","As the components are tool calls, they're already in the messages array in native format, so you don’t need to re-parse when you reload an old conversation. An example presentation-tool contract is illustrated below and in the reference repo.：https://github.com/anthropics/commerce-agents","The tradeoff is streaming granularity. Each top-level argument of a tool call buffers on the server for validation, so the sub-components of a presentation tool arrive in steps even with streaming on. This impacts perceived latency.","To get a token-level stream, set eager_input_streaming: true on the tool definition, which skips the buffering and with it the server-side schema guarantee.","In our evals, schema violations are very rare on Claude Sonnet-class models and up, but wrap the call in a retry for the cases where one slips through.","Presentation tools also give the agent a record of what's on screen. When a customer says \"the first hotel\" or \"the third one down on the left,\" the layout is in the messages array, in the arguments of the last presentation call.","For that to work, the arguments have to reflect the rendered layout, so structure them the way the UI is structured, as ordered rows and carousels rather than a flat list the client rearranges.","Latency matters in commerce, and consumer surfaces are the least forgiving. However, on agentic surfaces, what we have consistently seen move metrics like retention, engagement, and cart size is the quality of the outcome.","Whether the answer was relevant and the task actually completed was more critical to those metrics as compared to marginal latency gains.","So attack latency on two fronts. Minimize end-to-end latency through good engineering, and pair that with dropping perceived latency (since time spent watching an agent work reads as progress).","Every user has a latency budget, and the techniques below keep the agent inside it without spending intelligence to get there.","Task completion latency is the sum, over model turns, of time to last token plus tool processing. That gives you three levers to work towards: fewer turns, faster tools, and faster tokens. These levers sometimes compete, so the thing to minimize is the sum rather than any one of them.","Query complexity adds turns, and is generally out of your control. Model intelligence and relevant context help the agent get to task completion in fewer turns. Some of our key learnings in this area include:","Perceived latency is the time a user feels until the screen does something. It’s especially critical in consumer-facing use cases where any transaction friction impacts checkout rates and revenue. Two techniques shorten it without touching the model:","Prompt caching is your largest cost reduction candidate and commerce traffic is well-suited for it. Cached input token reads cost a tenth of fresh ones, and while cache-writes carry a premium of roughly 1.25x, a cached prefix pays for itself on its second use. In customer facing applications where volume is large, you have a unique opportunity to hit very high cache levels using the cheapest, default 5 minute cache expiration.","The best commerce deployments we've seen run at 90–99% cache hit rates, and that is the range to design for from the start. Our experience has shown cached token reads are also around 1.5 to 2x faster at ~100k tokens, with relatively linear scaling the more tokens there are.","Caching is prefix-based. A request reads from cache up to the first byte that differs from a previous request, so what matters is not just what is in the context but the order it is in. Think of a request as three segments, ordered by how often they change:","There are two implementation details to remember here. First, skills should be loaded as tool results rather than appended to the system prompt. The skill body then lands in the conversation prefix and is cached along with it.","Second, roll your breakpoints forward in each turn: a request allows a limited number of breakpoints, so move the newest one to the end of each user turn. Each round then reads the accumulated history, including long tool results such as search responses, from cache.","Model size and the effort setting：https://claude.com/blog/claude-model-and-effort-level-in-claude-code are the same tradeoff – intelligence against latency and cost – and you should choose both by measurement:","Measure cost per completed task rather than per model call, since a cheaper model that needs more turns, or fails more often, is not cheaper. When the result is close, and the cost fits your per-task economics and latency, choose intelligence. Quality is what drives adoption and retention, and allows for room to build for the next 6 months as models become better.","Lastly, we talk about what gets an agent through production: memory, safety, evals, and scaling the work across an organization.","The relationship and interactions you have with your customers matter. Memory is what lets an agent pick up where the last conversation left off instead of starting from nothing. A shopper who mentioned a nut allergy in March shouldn't have to repeat it in June, and a merchant who checks the same three campaigns every Monday shouldn't have to name them each time. Long-term memory, the facts that should survive across sessions, is a system you build and it has three parts: how facts are stored, how they are written, and how they are read."],"articleImages":[{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6903d222061abf091318fb82_423062049d4676b41d52b16068cbb5e21603190e-1000x1000.svg","alt":"","afterParagraph":0,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/4fa65b004ebab739.jpg"},{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6a97121e31e08caa3a0e6679_02653800.png","alt":"","afterParagraph":5,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/83873ff2924af2e0.png"},{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6a971accf6d9dcde640f87df_presentationtool.gif","alt":"","afterParagraph":38,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/13fb280533b2d4b9.gif"},{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6a971b4ebf113390b39a25b2_eagerdispatch.gif","alt":"","afterParagraph":49,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/88e9cdc43ee30fa0.gif"},{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6a971b28c43d0f061e80bc6c_perceivedlatency.gif","alt":"","afterParagraph":50,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/927644dd7cbb8182.gif"},{"sourceUrl":"https://cdn.prod.website-files.com/68a44d4040f98a4adf2207b6/6a970f654fd654f0e7990b95_c63ca0e7.png","alt":"","afterParagraph":53,"url":"/media/articles/cmtkcffah018zrog0zokk6s30/5e3a8f790f72f809.png"}],"mediaStatus":"ok","articleBodyZh":["使买卖更容易的代理的架构、延迟与成本技术以及评估实践。","在过去的一年中，我们与整个商业行业的团队——零售商、市场、旅游、娱乐和电信提供商——合作，使用Claude构建商业代理。","这些代理已投入生产，企业客户在使用它们时看到了更大的购物车和更高效的卖家操作。它们也共享一个简单的架构：Claude在一个代理循环中，配备了一组技能、工具和强大的评估套件。","本篇文章面向构建这些（或其他面向消费者）代理的工程师和工程领导者。第1部分涵盖架构，这是你只需决定一次的。第2部分涵盖延迟和成本。第3部分涵盖生产：内存、安全、评估以及在组织内扩展工作。","我们将商业代理定义为简化在线目录中买卖的代理。","有些代理面向消费者：它们搜索、比较、替代并组装订单。这可能是零售购物车、旅行行程、移动套餐变更或演出座位预订。有些代理面向企业：它们回答有关销售的问题、执行促销和活动、管理库存和定价。","核心架构是标准代理循环中的模型：https://www.anthropic.com/engineering/building-effective-agents：围绕目标进行推理、探索上下文、通过工具采取行动、通过技能学习流程、提出澄清问题并观察结果，直到目标完成。","前面没有意图路由器来划分对话，也没有一组特定领域的代理在其后。","商业代理必须涵盖多个类别和意图的广泛能力，这使得为每个领域创建一个子代理很诱人。","实际上，这被证明是次优的，因为商业对话是一个跨多个意图和轮次紧密耦合的会话，并且需要相当多的共享上下文。","在子代理架构中，协调器持有购物车或阶段性变更、用户偏好和对话历史记录。","每次交接给子代理都是一个状态丢失的操作，这通常会影响子代理的响应质量，进而影响整体响应质量。除此之外，每次交接可能会消耗几倍的令牌并增加数秒延迟。","这些领域也很少能完全分离。退货流程可能需要订单历史记录、当前购物车和产品目录，这意味着每个领域一个子代理的方法要么在各处重复访问这些数据，要么在任务中途进行交接。","随着模型变得更聪明，它们也能处理更长的上下文、更多技能和更多工具，因此现有放置规则背后的限制会随着每一代模型而放宽。","相反，代理技能：https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview 提供了类似的按领域模块化和上下文控制，而没有交接的成本，因为技能指令加载到已经保存全部历史记录的主代理中。","在我们对多个企业部署的比较中，具备技能的单一代理在质量上始终优于所有任务使用单一提示的设计和子代理设计，并且通常每个任务的成本和延迟都更低。","子代理确实有其作用的场合是，当协调器可以将其作为工具调用，用于狭窄或自包含的任务，这类任务将受益于其自身独立的上下文窗口。","一个常见的生产示例是深度研究子代理，其中子代理搜索和阅读文档、编写和运行代码、遍历数据模型，并可能遇到死胡同。所有工作都发生在一个或多个子代理内部，只有一个简明的答案返回给协调器。","另一种例外情况是领域已经有其专门构建的代理。如果你的药房或金融服务体验运行一个独立代理，该代理具有自己的合规界面，那么正确的做法可能是交接，让该代理接管任务，并通过自己的循环直接与用户交互，直到任务完成。","区别在于对话的所有权。交接使领域代理成为用户的对应方，而委托则保持协调器，将领域代理在单一回合中进出调用，并在每次交换中降低效果。","在决定将一组指令放入系统提示还是技能时，主要因素是代理需要它的频率。加载技能会消耗一个模型回合，因此代理在大多数回合中需要的任何内容通常都放在系统提示中。","然而，这取决于你的流量分布以及评估显示的代理行为。一个好的起点是，任何与三分之一或更多流量相关的内容，无论是在上线前预期还是在生产中观察到，都放在系统提示中，其余内容则放在技能中。","如果技能可以从你已有的信号预测，例如用户访问的页面，我们建议在第一次模型调用之前从程序框架注入，而跳过加载技能的额外回合。","关键指令，如安全和法律规则、品牌约束以及关键用户信息（如过敏信息），始终放在系统提示中。","对于商业代理而言，这意味着产品搜索放在提示中，因为几乎每个会话都会用到它，而技能则承担长尾功能。","在我们的参考实现中：https://github.com/anthropics/commerce-agents，购物代理的提示包含基础、购物车和结账语义，以及展示规则，以下技能涵盖其余部分：搜索发现、购买研究、规划目标、客户服务和记忆个性化。","商家代理的划分方式相同，其技能包括性能洞察（performance-insights）、目录列表（catalog-listings）、库存操作（inventory-operations）、定价促销（pricing-promotions）和营销活动（marketing-campaigns），每个技能对应一个运营领域。","我们关于为代理编写有效工具的文章：https://www.anthropic.com/engineering/writing-tools-for-agents 涵盖了工具设计的一般内容。在商业中，有两个要点最为重要：","在核心系统和逻辑之上构建代理工具。","一家商业公司已经拥有搜索和排序、购物车、偏好和档案存储、库存系统、促销和活动引擎、销售分析等，每个系统都经过多年调优，并获取模型永远无法看到的信号。","代理的工具应调用这些系统，而不是重新实现它们，工具边界是它们逻辑结束且模型判断开始的地方。","例如，当代理调用 search_products 时，结果应该已经排好序；它的工作是决定哪些结果符合用户的目标，展示多少，以及如何展示它们。","返回模型用于推理的字段，丢弃其余字段。每个搜索行的图片 URL 通常是罪魁祸首。","根据需要，在工具内部重塑原始响应，包括在数据不明确时追加下一步。","这在错误场景中尤为相关，因为模型受益于指令而不是错误代码。例如，添加一个错误指令“查询可用性时包含产品 ID”，而不是通用的 403。","大多数商务代理的响应是 UI 组件而不是文本，无论是产品轮播、行程、座位图还是图表。这意味着代理必须输出一个 schema 而不是文本。","团队有时会通过提示模型输出自定义标签并在客户端解析它们来开始。这种方法在界面扩展时会失效，因为：","经验证可行的模式是将每个 UI 组件制作成一个工具。模型调用 present_products、present_itinerary 或 present_plan_comparison 并传递类型化参数；你的服务器验证并丰富调用，然后发出事件；你的客户端进行渲染。","由于组件是工具调用，它们已经以原生格式存在于 messages 数组中，因此在重新加载旧对话时无需重新解析。示例展示工具契约如下，参考仓库链接：https://github.com/anthropics/commerce-agents","权衡点在于流式粒度。工具调用的每个顶级参数在服务器端缓存以进行验证，因此即使开启流式处理，展示工具的子组件也会分步骤到达。这会影响感知延迟。","要获得令牌级流式处理，在工具定义中设置 eager_input_streaming: true，这会跳过缓冲，从而也没有服务器端 schema 保证。","在我们的评估中，Claude Sonnet 类及以上模型的 schema 违规非常罕见，但为了应对偶尔出现的情况，请在调用中加入重试机制。","演示工具还为代理提供了屏幕内容的记录。当客户说“第一个酒店”或“左边第三个”时，布局信息在消息数组中，即最后一次演示调用的参数中。","为了使其工作，这些参数必须反映渲染后的布局，因此请按照UI的结构来组织它们，将其作为有序的行和轮播，而不是客户端重新排列的平面列表。","延迟在商业中很重要，而面向消费者的界面最不容忍。然而，在代理界面上，我们持续观察到能够提升保留率、参与度和购物车规模等指标的关键是结果的质量。","相比边际延迟的改善，答案是否相关以及任务是否真正完成，对这些指标更为关键。","因此从两个方面处理延迟。通过良好的工程实践来最小化端到端延迟，同时降低感知延迟（因为用户看到代理工作时，会将其视为进度）。","每个用户都有延迟预算，下列技术可以在不增加智能消耗的情况下，使代理保持在预算内。","任务完成延迟是模型轮次中“最后一个token的时间加上工具处理时间”的总和。这给了你三个可操作的杠杆：减少轮次、加快工具速度和加快token生成速度。这些杠杆有时会相互竞争，所以需要最小化的是总和，而不是单独的某一项。","查询复杂性会增加轮次，通常不在你的控制范围内。模型智能和相关上下文能帮助代理用更少轮次完成任务。我们在这方面的一些关键经验包括：","感知延迟是用户感觉到屏幕做出响应所需的时间。在面向消费者的使用场景中尤其关键，因为任何交易摩擦都会影响结账率和收入。有两种技术可以缩短它而无需修改模型：","提示缓存是您降低成本的最大候选，而商业流量非常适合使用它。缓存的输入令牌读取成本只有新读取的十分之一，而尽管缓存写入大约带来1.25倍的额外开销，一个缓存前缀在第二次使用时就能收回成本。在面对客户的应用中，如果访问量很大，您有一个独特的机会，使用最便宜的默认5分钟缓存过期时间达到非常高的缓存命中率。","我们看到的最好的商业部署缓存命中率在90%–99%之间，这也是从一开始就应该设计的范围。我们的经验显示，在大约10万令牌时，缓存令牌读取也快约1.5到2倍，且随着令牌数量增加，速度几乎线性增长。","缓存是基于前缀的。一个请求从缓存中读取，直到第一个与之前的请求不同的字节为止，因此重要的不仅是上下文中有什么，还包括它的顺序。将一个请求分成三个段，按变化频率排序：","这里有两个实现细节需要记住。首先，技能应作为工具结果加载，而不是附加到系统提示中。技能主体随后位于对话前缀中，并与其一起缓存。","其次，在每一轮中向前滚动你的断点：一个请求允许有限数量的断点，因此将最新的断点移动到每一轮用户回合的末尾。每一轮随后从缓存读取累积的历史记录，包括长的工具结果，如搜索响应。","模型大小和努力设置：https://claude.com/blog/claude-model-and-effort-level-in-claude-code 是相同的权衡——智能与延迟和成本之间的权衡——您应该通过测量来选择两者：","衡量每个完成任务的成本，而不是每次模型调用的成本，因为一个更便宜的模型如果需要更多轮次，或者失败率更高，并不意味着更便宜。当结果接近且成本符合每任务经济性和延迟要求时，选择智能。质量是驱动采用和用户留存的关键，并且在模型变得更好的未来6个月内为构建留出空间。","最后，我们讨论让代理顺利通过生产的因素：内存、安全性、评估，以及在组织中扩展工作。","你与客户的关系和互动很重要。记忆让代理能够从上一次对话结束的地方继续，而不是从零开始。三月份提到坚果过敏的购物者在六月不应该再重复这一信息，而每周一检查同三个活动的商家也不需要每次都列出它们。长期记忆，即应该在多次会话中保持的事实，是你需要构建的系统，它有三个部分：事实如何存储、如何写入以及如何读取。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Anthropic 发布面向电商智能体开发者的指南及参考实现，内容覆盖架构、延迟与成本、内存、安全、评估和组织扩展，并讨论购物端与商家端两类应用。","background":"文章称，Anthropic 过去一年与零售、市场平台、旅行、娱乐和电信团队合作构建基于 Claude 的电商智能体；部分企业客户在使用后观察到购物车变大和卖家运营效率提升。","viewpoint":"Aioga 判断：这份指南的重点是以单一智能体循环配合技能、工具和评估体系，避免在前端设置意图路由及在后端拆分多个领域智能体，体现了对架构简化和生产治理的重视。","implications":"可能影响：开发团队可能参考其单智能体架构、工具化界面组件、提示词缓存和安全规则设计，但文中所述客户表现不足以证明这些方法在不同业务环境中都能产生相同结果。","nextStep":"后续观察：需要关注开源参考实现 anthropics/commerce-agents 的具体设计、评估方法与安全约束，以及相关实践在不同在线目录和业务流程中的适用范围。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-03T16:32:35.097Z","sourceHash":"69291ab4f5721057","review":{"approved":true,"groundedness":96,"clarity":94,"duplicationRisk":8,"blockingIssues":[],"notes":["“工具化界面组件”“提示词缓存”和“安全规则设计”均可由来源摘要支持。","“前端设置意图路由及在后端拆分多个领域智能体”是对原文架构描述的概括，作为已明确标注的观点表达，不构成事实错误。","“客户表现不足以证明这些方法在不同业务环境中都能产生相同结果”属于合理限定，不是无来源断言。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","Claude：Blog（网页）"],"translations":{"zh-CN":{"title":"Anthropic 发布电商 Agent 架构与生产实践指南，并开源 commerce-agents 参考实现","summary":"Anthropic 发布电商 Agent 构建指南，基于与零售、旅游、电信等团队的落地经验，核心架构是单个 Claude 在标准 Agent 循环中配合技能与工具，而非按领域拆分子智能体，并开源了 anthropics/commerce-agents 参考实现，含购物与商家 Agent。","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic 发布电商 Agent 架构与生产实践指南，并开源 commerce-agents 参考实现 - Aioga AI资讯","description":"Anthropic 发布电商 Agent 构建指南，基于与零售、旅游、电信等团队的落地经验，核心架构是单个 Claude 在标准 Agent 循环中配合技能与工具，而非按领域拆分子智能体，并开源了 anthropics/commerce-agents 参考实现，含购物与商家 Agent。","url":"https://www.aioga.com/news/cmtkcffah018zrog0zokk6s30/","articleBody":["使买卖更容易的代理的架构、延迟与成本技术以及评估实践。","在过去的一年中，我们与整个商业行业的团队——零售商、市场、旅游、娱乐和电信提供商——合作，使用Claude构建商业代理。","这些代理已投入生产，企业客户在使用它们时看到了更大的购物车和更高效的卖家操作。它们也共享一个简单的架构：Claude在一个代理循环中，配备了一组技能、工具和强大的评估套件。","本篇文章面向构建这些（或其他面向消费者）代理的工程师和工程领导者。第1部分涵盖架构，这是你只需决定一次的。第2部分涵盖延迟和成本。第3部分涵盖生产：内存、安全、评估以及在组织内扩展工作。","我们将商业代理定义为简化在线目录中买卖的代理。","有些代理面向消费者：它们搜索、比较、替代并组装订单。这可能是零售购物车、旅行行程、移动套餐变更或演出座位预订。有些代理面向企业：它们回答有关销售的问题、执行促销和活动、管理库存和定价。","核心架构是标准代理循环中的模型：https://www.anthropic.com/engineering/building-effective-agents：围绕目标进行推理、探索上下文、通过工具采取行动、通过技能学习流程、提出澄清问题并观察结果，直到目标完成。","前面没有意图路由器来划分对话，也没有一组特定领域的代理在其后。","商业代理必须涵盖多个类别和意图的广泛能力，这使得为每个领域创建一个子代理很诱人。","实际上，这被证明是次优的，因为商业对话是一个跨多个意图和轮次紧密耦合的会话，并且需要相当多的共享上下文。","在子代理架构中，协调器持有购物车或阶段性变更、用户偏好和对话历史记录。","每次交接给子代理都是一个状态丢失的操作，这通常会影响子代理的响应质量，进而影响整体响应质量。除此之外，每次交接可能会消耗几倍的令牌并增加数秒延迟。","这些领域也很少能完全分离。退货流程可能需要订单历史记录、当前购物车和产品目录，这意味着每个领域一个子代理的方法要么在各处重复访问这些数据，要么在任务中途进行交接。","随着模型变得更聪明，它们也能处理更长的上下文、更多技能和更多工具，因此现有放置规则背后的限制会随着每一代模型而放宽。","相反，代理技能：https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview 提供了类似的按领域模块化和上下文控制，而没有交接的成本，因为技能指令加载到已经保存全部历史记录的主代理中。","在我们对多个企业部署的比较中，具备技能的单一代理在质量上始终优于所有任务使用单一提示的设计和子代理设计，并且通常每个任务的成本和延迟都更低。","子代理确实有其作用的场合是，当协调器可以将其作为工具调用，用于狭窄或自包含的任务，这类任务将受益于其自身独立的上下文窗口。","一个常见的生产示例是深度研究子代理，其中子代理搜索和阅读文档、编写和运行代码、遍历数据模型，并可能遇到死胡同。所有工作都发生在一个或多个子代理内部，只有一个简明的答案返回给协调器。","另一种例外情况是领域已经有其专门构建的代理。如果你的药房或金融服务体验运行一个独立代理，该代理具有自己的合规界面，那么正确的做法可能是交接，让该代理接管任务，并通过自己的循环直接与用户交互，直到任务完成。","区别在于对话的所有权。交接使领域代理成为用户的对应方，而委托则保持协调器，将领域代理在单一回合中进出调用，并在每次交换中降低效果。","在决定将一组指令放入系统提示还是技能时，主要因素是代理需要它的频率。加载技能会消耗一个模型回合，因此代理在大多数回合中需要的任何内容通常都放在系统提示中。","然而，这取决于你的流量分布以及评估显示的代理行为。一个好的起点是，任何与三分之一或更多流量相关的内容，无论是在上线前预期还是在生产中观察到，都放在系统提示中，其余内容则放在技能中。","如果技能可以从你已有的信号预测，例如用户访问的页面，我们建议在第一次模型调用之前从程序框架注入，而跳过加载技能的额外回合。","关键指令，如安全和法律规则、品牌约束以及关键用户信息（如过敏信息），始终放在系统提示中。","对于商业代理而言，这意味着产品搜索放在提示中，因为几乎每个会话都会用到它，而技能则承担长尾功能。","在我们的参考实现中：https://github.com/anthropics/commerce-agents，购物代理的提示包含基础、购物车和结账语义，以及展示规则，以下技能涵盖其余部分：搜索发现、购买研究、规划目标、客户服务和记忆个性化。","商家代理的划分方式相同，其技能包括性能洞察（performance-insights）、目录列表（catalog-listings）、库存操作（inventory-operations）、定价促销（pricing-promotions）和营销活动（marketing-campaigns），每个技能对应一个运营领域。","我们关于为代理编写有效工具的文章：https://www.anthropic.com/engineering/writing-tools-for-agents 涵盖了工具设计的一般内容。在商业中，有两个要点最为重要：","在核心系统和逻辑之上构建代理工具。","一家商业公司已经拥有搜索和排序、购物车、偏好和档案存储、库存系统、促销和活动引擎、销售分析等，每个系统都经过多年调优，并获取模型永远无法看到的信号。","代理的工具应调用这些系统，而不是重新实现它们，工具边界是它们逻辑结束且模型判断开始的地方。","例如，当代理调用 search_products 时，结果应该已经排好序；它的工作是决定哪些结果符合用户的目标，展示多少，以及如何展示它们。","返回模型用于推理的字段，丢弃其余字段。每个搜索行的图片 URL 通常是罪魁祸首。","根据需要，在工具内部重塑原始响应，包括在数据不明确时追加下一步。","这在错误场景中尤为相关，因为模型受益于指令而不是错误代码。例如，添加一个错误指令“查询可用性时包含产品 ID”，而不是通用的 403。","大多数商务代理的响应是 UI 组件而不是文本，无论是产品轮播、行程、座位图还是图表。这意味着代理必须输出一个 schema 而不是文本。","团队有时会通过提示模型输出自定义标签并在客户端解析它们来开始。这种方法在界面扩展时会失效，因为：","经验证可行的模式是将每个 UI 组件制作成一个工具。模型调用 present_products、present_itinerary 或 present_plan_comparison 并传递类型化参数；你的服务器验证并丰富调用，然后发出事件；你的客户端进行渲染。","由于组件是工具调用，它们已经以原生格式存在于 messages 数组中，因此在重新加载旧对话时无需重新解析。示例展示工具契约如下，参考仓库链接：https://github.com/anthropics/commerce-agents","权衡点在于流式粒度。工具调用的每个顶级参数在服务器端缓存以进行验证，因此即使开启流式处理，展示工具的子组件也会分步骤到达。这会影响感知延迟。","要获得令牌级流式处理，在工具定义中设置 eager_input_streaming: true，这会跳过缓冲，从而也没有服务器端 schema 保证。","在我们的评估中，Claude Sonnet 类及以上模型的 schema 违规非常罕见，但为了应对偶尔出现的情况，请在调用中加入重试机制。","演示工具还为代理提供了屏幕内容的记录。当客户说“第一个酒店”或“左边第三个”时，布局信息在消息数组中，即最后一次演示调用的参数中。","为了使其工作，这些参数必须反映渲染后的布局，因此请按照UI的结构来组织它们，将其作为有序的行和轮播，而不是客户端重新排列的平面列表。","延迟在商业中很重要，而面向消费者的界面最不容忍。然而，在代理界面上，我们持续观察到能够提升保留率、参与度和购物车规模等指标的关键是结果的质量。","相比边际延迟的改善，答案是否相关以及任务是否真正完成，对这些指标更为关键。","因此从两个方面处理延迟。通过良好的工程实践来最小化端到端延迟，同时降低感知延迟（因为用户看到代理工作时，会将其视为进度）。","每个用户都有延迟预算，下列技术可以在不增加智能消耗的情况下，使代理保持在预算内。","任务完成延迟是模型轮次中“最后一个token的时间加上工具处理时间”的总和。这给了你三个可操作的杠杆：减少轮次、加快工具速度和加快token生成速度。这些杠杆有时会相互竞争，所以需要最小化的是总和，而不是单独的某一项。","查询复杂性会增加轮次，通常不在你的控制范围内。模型智能和相关上下文能帮助代理用更少轮次完成任务。我们在这方面的一些关键经验包括：","感知延迟是用户感觉到屏幕做出响应所需的时间。在面向消费者的使用场景中尤其关键，因为任何交易摩擦都会影响结账率和收入。有两种技术可以缩短它而无需修改模型：","提示缓存是您降低成本的最大候选，而商业流量非常适合使用它。缓存的输入令牌读取成本只有新读取的十分之一，而尽管缓存写入大约带来1.25倍的额外开销，一个缓存前缀在第二次使用时就能收回成本。在面对客户的应用中，如果访问量很大，您有一个独特的机会，使用最便宜的默认5分钟缓存过期时间达到非常高的缓存命中率。","我们看到的最好的商业部署缓存命中率在90%–99%之间，这也是从一开始就应该设计的范围。我们的经验显示，在大约10万令牌时，缓存令牌读取也快约1.5到2倍，且随着令牌数量增加，速度几乎线性增长。","缓存是基于前缀的。一个请求从缓存中读取，直到第一个与之前的请求不同的字节为止，因此重要的不仅是上下文中有什么，还包括它的顺序。将一个请求分成三个段，按变化频率排序：","这里有两个实现细节需要记住。首先，技能应作为工具结果加载，而不是附加到系统提示中。技能主体随后位于对话前缀中，并与其一起缓存。","其次，在每一轮中向前滚动你的断点：一个请求允许有限数量的断点，因此将最新的断点移动到每一轮用户回合的末尾。每一轮随后从缓存读取累积的历史记录，包括长的工具结果，如搜索响应。","模型大小和努力设置：https://claude.com/blog/claude-model-and-effort-level-in-claude-code 是相同的权衡——智能与延迟和成本之间的权衡——您应该通过测量来选择两者：","衡量每个完成任务的成本，而不是每次模型调用的成本，因为一个更便宜的模型如果需要更多轮次，或者失败率更高，并不意味着更便宜。当结果接近且成本符合每任务经济性和延迟要求时，选择智能。质量是驱动采用和用户留存的关键，并且在模型变得更好的未来6个月内为构建留出空间。","最后，我们讨论让代理顺利通过生产的因素：内存、安全性、评估，以及在组织中扩展工作。","你与客户的关系和互动很重要。记忆让代理能够从上一次对话结束的地方继续，而不是从零开始。三月份提到坚果过敏的购物者在六月不应该再重复这一信息，而每周一检查同三个活动的商家也不需要每次都列出它们。长期记忆，即应该在多次会话中保持的事实，是你需要构建的系统，它有三个部分：事实如何存储、如何写入以及如何读取。"]},"en":{"title":"Anthropic releases Claude e-commerce intelligent agent construction guide and reference implementation","summary":"Anthropic released a guide for building e-commerce agents, summarizing practices in three areas: architecture, latency and cost optimization, and production operations. It involves two types of agents: shopping and merchant agents. After enterprise clients deploy them, shopping carts increase in size and seller operational efficiency improves. Key recommendations include using a single agent with skills rather than sub-agents, turning UI components into tools, achieving a 90-99% hit rate with prompt caching, enforcing safety rules in the harness, and open-sourcing the reference implementation anthropics/commerce-agents.","category":"Industry","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic releases Claude e-commerce intelligent agent construction guide and reference implementation - Aioga AI News","description":"Anthropic released a guide for building e-commerce agents, summarizing practices in three areas: architecture, latency and cost optimization, and production operations. It involves...","url":"https://www.aioga.com/en/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:23:22.371Z"},"ja":{"title":"Anthropic、ClaudeのEC向けインテリジェントエージェント構築ガイドおよび参考実装を公開","summary":"Anthropicは、Eコマース向けインテリジェントエージェント構築のガイドを公開し、アーキテクチャ、遅延とコストの最適化、運用保守の3つの実践をまとめています。対象はショッピングおよび商家の2種類のエージェントで、企業顧客が導入した後には、カートの拡大や販売者の運営効率の向上が見られました。主要な提案には、サブエージェントではなく単一のエージェントにスキルを追加すること、UIコンポーネントをツール化すること、プロンプトキャッシュで90〜99％の命中率を実現すること、harness内で安全ルールを強制すること、そしてリファレンス実装 anthropics/commerce-agents をオープンソース化することが含まれます。","category":"業界動向","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic、ClaudeのEC向けインテリジェントエージェント構築ガイドおよび参考実装を公開 - Aioga AIニュース","description":"Anthropicは、Eコマース向けインテリジェントエージェント構築のガイドを公開し、アーキテクチャ、遅延とコストの最適化、運用保守の3つの実践をまとめています。対象はショッピングおよび商家の2種類のエージェントで、企業顧客が導入した後には、カートの拡大や販売者の運営効率の向上が見られました。主要な提案には、サブエージェントではなく単一のエージェントにスキル...","url":"https://www.aioga.com/ja/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:23:29.779Z"},"ko":{"title":"Anthropic, Claude 전자상거래 인공지능 구축 가이드 및 참조 구현 발표","summary":"Anthropic는 전자상거래 인텔리전트 에이전트 구축 가이드를 발표했으며, 아키텍처, 지연 및 비용 최적화, 운영 관리의 세 가지 실습을 정리했습니다. 이 가이드는 쇼핑과 판매자 두 가지 유형의 에이전트에 관련되며, 기업 고객이 배포한 후에는 장바구니가 커지고 판매자 운영 효율성이 향상되었습니다. 핵심 권장 사항에는 하위 에이전트 대신 단일 에이전트에 스킬을 추가하고, UI 구성 요소를 도구로 만들며, 프롬프트 캐싱을 통해 90-99% 적중률을 달성하고, 하네스(harness)에서 안전 규칙을 강제하며, 참고 구현인 anthropics/commerce-agents를 오픈소스화하는 것이 포함됩니다.","category":"업계 동향","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic, Claude 전자상거래 인공지능 구축 가이드 및 참조 구현 발표 - Aioga AI 뉴스","description":"Anthropic는 전자상거래 인텔리전트 에이전트 구축 가이드를 발표했으며, 아키텍처, 지연 및 비용 최적화, 운영 관리의 세 가지 실습을 정리했습니다. 이 가이드는 쇼핑과 판매자 두 가지 유형의 에이전트에 관련되며, 기업 고객이 배포한 후에는 장바구니가 커지고 판매자 운영 효율성이 향상되었습니다. 핵심 권장 사항에는 하...","url":"https://www.aioga.com/ko/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:24:17.929Z"},"es":{"title":"Anthropic publica la guía de construcción del agente inteligente de comercio electrónico Claude y su implementación de referencia","summary":"Anthropic publicó una guía para construir agentes inteligentes de comercio electrónico, resumiendo prácticas en tres partes: arquitectura, optimización de latencia y costos, y operaciones de producción, involucrando dos tipos de agentes: de compras y de comerciantes. Tras la implementación por parte de clientes empresariales, se observó un aumento en el tamaño del carrito de compras y una mejora en la eficiencia operativa de los vendedores. Las recomendaciones clave incluyen usar un solo agente con habilidades en lugar de subagentes, diseñar los componentes de la interfaz de usuario como herramientas, usar caché de prompts para lograr una tasa de aciertos del 90-99%, imponer reglas de seguridad en el harness y publicar como código abierto la implementación de referencia anthropics/commerce-agents.","category":"Industria","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic publica la guía de construcción del agente inteligente de comercio electrónico Claude y su implementación de referencia - Aioga Noticias de IA","description":"Anthropic publicó una guía para construir agentes inteligentes de comercio electrónico, resumiendo prácticas en tres partes: arquitectura, optimización de latencia y costos, y oper...","url":"https://www.aioga.com/es/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:24:17.982Z"},"fr":{"title":"Anthropic publie le guide de construction et la mise en œuvre de référence de l'agent intelligent Claude pour le commerce électronique","summary":"Anthropic a publié un guide pour la construction d’agents intelligents pour le commerce électronique, résumant trois parties de pratiques : architecture, optimisation de la latence et des coûts, et exploitation en production. Le guide couvre deux types d’agents : consommateurs et commerçants. Après déploiement chez des clients entreprises, on a constaté une augmentation du panier moyen et une amélioration de l’efficacité opérationnelle des vendeurs. Les recommandations principales incluent l’utilisation d’un agent unique avec des compétences plutôt que des sous-agents, la transformation des composants UI en outils, l’utilisation d’un cache de prompts pour atteindre un taux de succès de 90 à 99 %, l’application obligatoire des règles de sécurité dans le harness, et l’ouverture du code de référence anthropics/commerce-agents.","category":"Industrie","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic publie le guide de construction et la mise en œuvre de référence de l'agent intelligent Claude pour le commerce électronique - Aioga Actualités IA","description":"Anthropic a publié un guide pour la construction d’agents intelligents pour le commerce électronique, résumant trois parties de pratiques : architecture, optimisation de la latence...","url":"https://www.aioga.com/fr/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:25:07.472Z"},"de":{"title":"Anthropic veröffentlicht Leitfaden und Referenzimplementierung für den Aufbau des Claude E-Commerce-Intelligenzsystems","summary":"Anthropic hat einen Leitfaden zum Aufbau von E-Commerce-Intelligent-Agenten veröffentlicht, der die Praktiken in den drei Bereichen Architektur, Latenz- und Kostenoptimierung sowie Produktionsbetrieb zusammenfasst. Er bezieht sich auf zwei Arten von Agenten: Einkaufs- und Anbieter-Agenten. Nach der Implementierung bei Unternehmenskunden stiegen der Einkaufswagenwert und die Effizienz der Anbieteroperationen. Die Kernempfehlungen umfassen die Verwendung eines einzelnen Agenten mit Fähigkeiten anstelle von Unter-Agenten, die Gestaltung von UI-Komponenten als Werkzeuge, die Nutzung von Prompt-Caching zur Erreichung einer Trefferquote von 90-99 %, die Durchsetzung von Sicherheitsregeln im Harness und die Veröffentlichung einer Referenzimplementierung unter anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic veröffentlicht Leitfaden und Referenzimplementierung für den Aufbau des Claude E-Commerce-Intelligenzsystems - Aioga KI-News","description":"Anthropic hat einen Leitfaden zum Aufbau von E-Commerce-Intelligent-Agenten veröffentlicht, der die Praktiken in den drei Bereichen Architektur, Latenz- und Kostenoptimierung sowie...","url":"https://www.aioga.com/de/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:25:05.034Z"},"pt-BR":{"title":"Anthropic lança guia de construção do agente inteligente Claude para e-commerce e implementação de referência","summary":"Anthropic publicou um guia para construir agentes inteligentes de e-commerce, resumindo práticas em três partes: arquitetura, otimização de latência e custo, e operação de produção, envolvendo dois tipos de agentes, para compras e para comerciantes. Após a implementação pelos clientes corporativos, observou-se aumento no tamanho do carrinho de compras e melhoria na eficiência operacional dos vendedores. As recomendações principais incluem usar um único agente com habilidades em vez de subagentes, transformar os componentes de UI em ferramentas, usar cache de prompts para alcançar 90-99% de taxa de acerto, impor regras de segurança no harness, e abrir o código de implementação de referência anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic lança guia de construção do agente inteligente Claude para e-commerce e implementação de referência - Aioga Notícias de IA","description":"Anthropic publicou um guia para construir agentes inteligentes de e-commerce, resumindo práticas em três partes: arquitetura, otimização de latência e custo, e operação de produção...","url":"https://www.aioga.com/pt-BR/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:25:53.269Z"},"ru":{"title":"Anthropic 发布 Claude коммерческий интеллектуальный агент: руководство по созданию и справочная реализация","summary":"Anthropic опубликовала руководство по созданию интеллектуальных агентов для электронной коммерции, суммируя практики в трех областях: архитектура, оптимизация задержек и затрат, эксплуатация и поддержка. Руководство охватывает два типа агентов — для покупателей и для продавцов. После развертывания у корпоративных клиентов наблюдалось увеличение размера корзины и повышение эффективности работы продавцов. Ключевые рекомендации включают использование одного агента с навыками вместо дочерних агентов, создание UI-компонентов в виде инструментов, использование кэширования подсказок для достижения точности 90-99%, принудительное соблюдение правил безопасности в harness и открытый исходный код эталонной реализации anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic 发布 Claude коммерческий интеллектуальный агент: руководство по созданию и справочная реализация - Aioga Новости ИИ","description":"Anthropic опубликовала руководство по созданию интеллектуальных агентов для электронной коммерции, суммируя практики в трех областях: архитектура, оптимизация задержек и затрат, эк...","url":"https://www.aioga.com/ru/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:26:01.796Z"},"ar":{"title":"أعلنت شركة Anthropc عن دليل بناء وكيل ذكاء اصطناعي للتجارة الإلكترونية Claude والتنفيذ المرجعي","summary":"أصدرت Anthropic دليلاً لبناء وكلاء ذكيين للتجارة الإلكترونية، يلخص الممارسات في ثلاثة أجزاء: البنية، تحسين التأخير والتكلفة، وتشغيل وصيانة الإنتاج، ويشمل وكلاء للتسوق والبائعين. بعد نشره لدى العملاء من الشركات، لوحظ زيادة حجم سلة التسوق وتحسين كفاءة تشغيل البائعين. تشمل التوصيات الأساسية استخدام وكيل ذكي واحد مع مهارات بدلاً من وكلاء فرعيين، تصميم مكونات واجهة المستخدم كأدوات، استخدام ذاكرة التخزين المؤقت للإرشادات لتحقيق نسبة نجاح 90-99٪، فرض قواعد أمان في الـ harness، وإتاحة التطبيق المرجعي مفتوح المصدر anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"أعلنت شركة Anthropc عن دليل بناء وكيل ذكاء اصطناعي للتجارة الإلكترونية Claude والتنفيذ المرجعي - Aioga أخبار الذكاء الاصطناعي","description":"أصدرت Anthropic دليلاً لبناء وكلاء ذكيين للتجارة الإلكترونية، يلخص الممارسات في ثلاثة أجزاء: البنية، تحسين التأخير والتكلفة، وتشغيل وصيانة الإنتاج، ويشمل وكلاء للتسوق والبائعين. بع...","url":"https://www.aioga.com/ar/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:26:56.154Z"},"hi":{"title":"Anthropic ने Claude ई-कॉमर्स इंटेलिजेंट एजेंट निर्माण गाइड और संदर्भ कार्यान्वयन जारी किया","summary":"Anthropic ने ई-कॉमर्स स्मार्ट एजेंट बनाने के指南 को जारी किया, जिसमें आर्किटेक्चर, लेटेंसी और लागत अनुकूलन, और उत्पादन संचालन के तीन हिस्सों के अभ्यास का सारांश दिया गया है। इसमें शॉपिंग और मर्चेंट दोनों प्रकार के एजेंट शामिल हैं। जब व्यावसायिक ग्राहक इसे तैनात करते हैं, तो शॉपिंग कार्ट बड़ा होता है और विक्रेता की संचालन दक्षता बढ़ जाती है। मुख्य सुझावों में शामिल हैं: एकल एजेंट के साथ कौशल का उपयोग करना न कि उप-एजेंटों का, UI घटकों को टूल के रूप में बनाना, प्रॉम्प्ट कैशिंग का उपयोग करके 90-99% हिट रेट प्राप्त करना, हार्नेस में सुरक्षा नियम लागू करना, और संदर्भ कार्यान्वयन anthropics/commerce-agents को ओपन सोर्स करना।","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic ने Claude ई-कॉमर्स इंटेलिजेंट एजेंट निर्माण गाइड और संदर्भ कार्यान्वयन जारी किया - Aioga AI समाचार","description":"Anthropic ने ई-कॉमर्स स्मार्ट एजेंट बनाने के指南 को जारी किया, जिसमें आर्किटेक्चर, लेटेंसी और लागत अनुकूलन, और उत्पादन संचालन के तीन हिस्सों के अभ्यास का सारांश दिया गया है। इसमें शॉ...","url":"https://www.aioga.com/hi/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:26:57.991Z"},"it":{"title":"Anthropic pubblica la guida alla costruzione e l'implementazione di riferimento dell'agente intelligente per il commercio elettronico Claude","summary":"Anthropic ha pubblicato una guida per costruire agenti intelligenti per l'e-commerce, riassumendo le pratiche in tre parti: architettura, ottimizzazione di latenza e costi, e operazioni di produzione, coinvolgendo due tipi di agenti intelligenti, per lo shopping e per i commercianti. Dopo il dispiegamento ai clienti aziendali, si è osservato un aumento del carrello della spesa e un miglioramento dell'efficienza operativa dei venditori. I suggerimenti principali includono l'uso di un singolo agente con competenze piuttosto che sotto-agenti, trasformare i componenti UI in strumenti, utilizzare la memorizzazione nella cache dei prompt per ottenere un tasso di successo del 90-99%, applicare regole di sicurezza obbligatorie nell'harness, e rendere open source l'implementazione di riferimento anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic pubblica la guida alla costruzione e l'implementazione di riferimento dell'agente intelligente per il commercio elettronico Claude - Aioga Notizie IA","description":"Anthropic ha pubblicato una guida per costruire agenti intelligenti per l'e-commerce, riassumendo le pratiche in tre parti: architettura, ottimizzazione di latenza e costi, e opera...","url":"https://www.aioga.com/it/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:27:48.009Z"},"nl":{"title":"Anthropic publiceert Claude-gids voor het bouwen van e-commerce intelligentie en referentie-implementatie","summary":"Anthropic heeft een gids uitgebracht voor het bouwen van slimme e-commerce-agenten, waarin praktijkervaringen worden samengevat op het gebied van architectuur, vertraging en kostenoptimalisatie, en productie en operationele werking. Het betreft twee typen agenten: voor winkelen en voor verkopers. Na implementatie bij zakelijke klanten leidde dit tot grotere winkelwagentjes en een hogere operationele efficiëntie voor verkopers. Belangrijke aanbevelingen zijn onder andere het gebruik van een enkele agent met vaardigheden in plaats van subagenten, het maken van UI-componenten als tools, het behalen van een trefferpercentage van 90-99% via promptcaching, het afdwingen van veiligheidsregels in de harness, en het openstellen van referentie-implementaties via anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic publiceert Claude-gids voor het bouwen van e-commerce intelligentie en referentie-implementatie - Aioga AI-nieuws","description":"Anthropic heeft een gids uitgebracht voor het bouwen van slimme e-commerce-agenten, waarin praktijkervaringen worden samengevat op het gebied van architectuur, vertraging en kosten...","url":"https://www.aioga.com/nl/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:27:46.037Z"},"tr":{"title":"Anthropic, Claude e-ticaret yapay zekâ ajanı oluşturma kılavuzunu ve referans uygulamayı yayımladı","summary":"Anthropic, e-ticaret zekâ birimleri oluşturma rehberi yayımladı; bu rehber, mimari, gecikme ve maliyet optimizasyonu ile üretim operasyonu olmak üzere üç bölümden oluşan uygulamaları özetliyor ve alışveriş ile satıcı olmak üzere iki tür zekâ birimini kapsıyor. Kurumsal müşteriler dağıttıktan sonra alışveriş sepetinin büyümesi ve satıcı operasyon verimliliğinin artması gözlemlendi. Temel öneriler arasında alt zekâ birimleri yerine tek bir zekâ birimi ve yetenekler kullanmak, UI bileşenlerini araç olarak tasarlamak, ipucu önbellekleme ile %90-99 isabet oranı sağlamak, harness içinde güvenlik kurallarını zorunlu kılmak ve referans uygulamayı anthropics/commerce-agents olarak açık kaynak yapmak yer alıyor.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic, Claude e-ticaret yapay zekâ ajanı oluşturma kılavuzunu ve referans uygulamayı yayımladı - Aioga AI Haberleri","description":"Anthropic, e-ticaret zekâ birimleri oluşturma rehberi yayımladı; bu rehber, mimari, gecikme ve maliyet optimizasyonu ile üretim operasyonu olmak üzere üç bölümden oluşan uygulamala...","url":"https://www.aioga.com/tr/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:28:40.460Z"},"vi":{"title":"Anthropic phát hành hướng dẫn xây dựng và triển khai tham khảo Claude cho thương mại điện tử","summary":"Anthropic phát hành hướng dẫn xây dựng tác nhân thông minh cho thương mại điện tử, tổng kết thực tiễn về kiến trúc, tối ưu độ trễ và chi phí, vận hành sản xuất, liên quan đến hai loại tác nhân là mua sắm và thương nhân. Sau khi triển khai cho khách hàng doanh nghiệp, xảy ra hiện tượng giỏ hàng to hơn và hiệu quả vận hành của người bán tăng. Các khuyến nghị cốt lõi bao gồm sử dụng một tác nhân duy nhất kèm kỹ năng thay vì các tác nhân con, biến các thành phần UI thành công cụ, sử dụng bộ nhớ đệm prompt để đạt tỷ lệ trúng 90-99%, thực thi các quy tắc an toàn trong harness, đồng thời mở mã nguồn triển khai tham khảo anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic phát hành hướng dẫn xây dựng và triển khai tham khảo Claude cho thương mại điện tử - Tin tức AI Aioga","description":"Anthropic phát hành hướng dẫn xây dựng tác nhân thông minh cho thương mại điện tử, tổng kết thực tiễn về kiến trúc, tối ưu độ trễ và chi phí, vận hành sản xuất, liên quan đến hai l...","url":"https://www.aioga.com/vi/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:28:40.632Z"},"id":{"title":"Anthropic merilis panduan pembangunan dan implementasi referensi Claude untuk agen cerdas e-commerce","summary":"Anthropic merilis panduan untuk membangun agen cerdas e-commerce, merangkum praktik dalam tiga bagian: arsitektur, optimisasi latensi dan biaya, serta operasi produksi, mencakup dua jenis agen cerdas yaitu untuk belanja dan pedagang. Setelah diterapkan oleh klien perusahaan, terlihat peningkatan ukuran keranjang belanja dan efisiensi operasional penjual. Saran inti termasuk menggunakan satu agen cerdas dengan kemampuan tambahan daripada agen cerdas anak, membuat komponen UI menjadi alat, menggunakan cache prompt untuk mencapai tingkat keberhasilan 90-99%, menegakkan aturan keamanan di harness, dan membuka implementasi referensi di anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic merilis panduan pembangunan dan implementasi referensi Claude untuk agen cerdas e-commerce - Berita AI Aioga","description":"Anthropic merilis panduan untuk membangun agen cerdas e-commerce, merangkum praktik dalam tiga bagian: arsitektur, optimisasi latensi dan biaya, serta operasi produksi, mencakup du...","url":"https://www.aioga.com/id/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:29:27.771Z"},"th":{"title":"Anthropic 发布 Claude อีคอมเมิร์ซคู่มือการสร้างปัญญาประดิษฐ์และการนำไปใช้อ้างอิง","summary":"Anthropic เผยแพร่คำแนะนำในการสร้างเอเจนต์อัจฉริยะด้านอีคอมเมิร์ซ สรุปการปฏิบัติในสามส่วนคือ สถาปัตยกรรม การปรับค่าเวลาแฝงและต้นทุน และการผลิตและการดำเนินงาน ครอบคลุมเอเจนต์สองประเภทคือ ผู้ซื้อและผู้ขาย หลังจากที่ลูกค้าองค์กรนำไปปรับใช้ พบว่าตะกร้าสินค้าใหญ่ขึ้นและประสิทธิภาพการดำเนินงานของผู้ขายสูงขึ้น คำแนะนำหลักรวมถึงการใช้เอเจนต์เดียวพร้อมทักษะแทนเอเจนต์ย่อย การทำคอมโพเนนต์ UI ให้เป็นเครื่องมือ การใช้แคชของ prompt เพื่อให้มีอัตราการตอบกลับ 90-99% และบังคับใช้กฎความปลอดภัยใน harness พร้อมเปิดซอร์สตัวอย่าง implementation ที่ anthropics/commerce-agents","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic 发布 Claude อีคอมเมิร์ซคู่มือการสร้างปัญญาประดิษฐ์และการนำไปใช้อ้างอิง - ข่าว AI Aioga","description":"Anthropic เผยแพร่คำแนะนำในการสร้างเอเจนต์อัจฉริยะด้านอีคอมเมิร์ซ สรุปการปฏิบัติในสามส่วนคือ สถาปัตยกรรม การปรับค่าเวลาแฝงและต้นทุน และการผลิตและการดำเนินงาน ครอบคลุมเอเจนต์สองประเภ...","url":"https://www.aioga.com/th/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:29:31.626Z"},"pl":{"title":"Anthropic opublikował przewodnik budowy inteligentnego agenta handlu elektronicznego Claude oraz przykładową implementację","summary":"Anthropic opublikował przewodnik dotyczący budowy inteligentnych agentów dla e-commerce, podsumowując praktyki w trzech obszarach: architektura, optymalizacja opóźnień i kosztów oraz produkcja i utrzymanie. Dotyczy to dwóch typów agentów: dla zakupów i dla sprzedawców. Po wdrożeniu u klientów korporacyjnych zaobserwowano zwiększenie wielkości koszyka zakupowego oraz wzrost efektywności operacyjnej sprzedawców. Kluczowe zalecenia obejmują użycie pojedynczego agenta ze zdolnościami zamiast podagentów, tworzenie komponentów UI jako narzędzi, osiąganie 90–99% trafności za pomocą cache’owania promptów, egzekwowanie zasad bezpieczeństwa w harness oraz udostępnienie referencyjnej implementacji open source anthropics/commerce-agents.","category":"行业动态","source":"Claude：Blog（网页）","aggregationSource":"Claude：Blog（网页）","pageTitle":"Anthropic opublikował przewodnik budowy inteligentnego agenta handlu elektronicznego Claude oraz przykładową implementację - Aioga Wiadomości AI","description":"Anthropic opublikował przewodnik dotyczący budowy inteligentnych agentów dla e-commerce, podsumowując praktyki w trzech obszarach: architektura, optymalizacja opóźnień i kosztów or...","url":"https://www.aioga.com/pl/news/cmtkcffah018zrog0zokk6s30/","contentTranslated":true,"sourceHash":"d1f3e6644d6a1fa3","translatedAt":"2026-09-03T16:30:18.002Z"}},"evidenceTier":"verified-news","reviewStatus":"editorial-selected","indexable":true,"editorialCover":""}}