文 | 伯虎财经，作者 | 秦三钱
今年双11，购物车里多了一位特殊的参与者。
它可以替消费者筛选商品、比价、叠加优惠券，甚至主动申请价保退款；也能替商家盯库存、分析投放效果、批量生成营销素材。过去需要买卖双方反复操作的复杂繁琐流程，如今正交由 AI 接管。
10月13日，阿里推出个人智能助理“千问Go！”，将购物保价、比价和生活事务办理纳入AI能力圈。次日，千问办公电商版发布，试图用一个AI工作台统一管理商家跨平台的生意。
几乎同一时间，豆包手机助手却遭遇到一场集体抵制。
10月14日，豆包手机助手公布《屏幕自动化操作声明协议》（SAEP）最新进展，超过150款应用明确拒绝被 AI 自动化操作。淘宝、支付宝、美团、拼多多等高频应用均列其中，京东则暂未出现。
一边是电商平台争相把AI嵌入购物链路，另一边则是主流App坚决防范外部A 插手自身的交易流程。
AI 正在重塑电商，但各家公司试图占据的要塞截然不同。
对千问而言，电商是通向C端个人AI助理与B端企业级Agent的最佳切口；对豆包而言，购物是验证跨应用任务执行能力的核心场景；而对京东和拼多多来说，AI的价值始终围绕选品、转化、经营效率等传统电商指标展开。
相似的能力背后，藏着截然不同的商业算盘。这场围绕购物车展开的新博弈，正在双11期间彻底浮出水面。
01 阿里的两本账
阿里今年双11的AI布局，沿着用户与商家两端同步展开。
在用户侧，千问的角色正在从“购物顾问”升级为“购物代理人”。
今年5月，千问App 与淘宝全面打通，接入淘天超40亿商品，用户通过自然语言即可完成发现、比较、下单及售后全流程。淘宝App内部也上线了千问AI购物助手，主打AI试穿与优惠统筹。
淘宝与天猫的业态更多集中于货架，用户需要将需求简化为关键词进行搜索，发掘并比较出心仪的商品。如今，千问试图重构交互逻辑，例如买小户型家具，用户只需描述居住空间与预算，AI便能自动完成条件匹配与筛选。
10月13日测试的“千问Go！”则更进一步，开始介入购物保价、航班理赔、汽车保养询价等具体事务。从“帮消费者找商品”到“替消费者办事”，千问开始从智能搜索框跃迁为用户表达消费需求的第一入口。
这届双 11，顺理成章成为淘天新的“购物代理人”的考场，考核指标是订单转化率、客单价与复购率。双11期间规则繁复，AI替用户计算最优解，能直接降低算价成本，挽回因繁琐而放弃的订单。
另一方面，早已被验证的购物节，是中文互联网的流量洪峰。高频的电商交易，同样是培养超级AI入口的理想所在，考核指标则是用户将日常事务交由 AI 处理的留存习惯。
阿里的底气在于生态闭环。淘宝、支付宝、高德、飞猪构成了完整的履约体系。千问帮淘宝做成交易，淘宝反向推动千问成为全能的 AI 办事入口。
在商家端，千问办公电商版则试图将这一逻辑复制到企业服务领域。
10月14日发布的千问办公电商版，提供跨平台经营看板、趋势洞察、素材生成及跨平台ROI优化等功能，并接入生意参谋、千牛、有赞等经营工具。发布会数据显示，电商行业用户已占其付费用户的20%以上。
多平台经营是当下商家的常态，但订单、库存与广告数据分散在不同后台。千问办公试图成为独立第三方电商 Agent，将散落的数据统一收拢至同一个AI工作台中。
这一定位的野心在于，如果商家习惯将各平台的报表、营销方案与库存策略交给千问办公，它所沉淀的经营上下文与工作流，将转化为极其稳固的B端软件壁垒。
尽管现阶段千问办公的核心能力仍停留在数据整合与决策辅助，具体投放仍受限于API权限与人工确认，但其战略意图已非常清晰。
阿里不仅要用 AI 让淘宝卖出更多商品，更要借电商场景，让千问扎根于消费者与企业的日常工作流中。
02 豆包撞上了谁的墙？
如果说千问的购物能力建立在阿里自有生态之内，那么豆包手机助手试图解决的，是另一个终极难题。
用户能否跳过 App，直接向AI下达任务？
今年9月，依托手机系统权限、API 接口与 GUI（图形用户界面）自动化技术，努比亚豆包二代手机助手开始尝试理解用户指令，模拟人工点击、滑动与输入，跨应用执行搜题、比价和下单。
这种GUI自动化模式大大降低了调用现有App的门槛，也将Agent推到了生态对抗的前沿。
模拟人工操作虽能为用户省去繁琐步骤，却极易触及平台的账户权限、支付安全与风控红线。更深层的冲突在于对商业流量入口的再分配。
传统电商的商业模式建立在“搜索—推荐—浏览—广告”的链路之上，平台通过曝光和流量分配实现变现。一旦消费者将需求全权托管给外部AI，商品筛选与比较将在App之外完成。
当外部 AI 重新排序商品、分发流量时，电商平台原有的广告变现机制与推荐算法便面临失效危险。平台或许依然提供支付与履约，却失去了最核心的用户注意力与流量掌控力。这正是豆包手机助手遭受集体抵制的深层原因。
如前所述，截至10月14日，拒绝豆包SAEP协议的名单已增加至 150 余款，囊括淘宝、支付宝、美团、拼多多、微信等巨头。
需要厘清的是，平台抵制的是外部助手对自身应用的自动化越权操作，而非技术本身。 淘宝有自身的 AI 助手，拼多多有站内 AI 搜索，各家欢迎AI提升自身平台的交易效率，但坚决防范外部Agent越过规则侵蚀用户关系。
对豆包而言，购物是证明自身跨应用Agent调配能力的试金石；但对交易平台而言，这种能力一旦规模化，App将沦为纯粹的后台履约工具。AI越想成为独立的超级入口，受到的生态阻力就越大。
这也迫使行业思考另一个方向：若不争夺独立入口，仅让AI赋能现有的电商交易，阻力会不会小得多？
03 京东与拼多多，另一笔交易账
在豆包的拒绝名单中，京东是一个微妙的例外。
截至10月14日，京东系应用未明确列入拒绝名单。在商业逻辑上，这暗示了一种选择的可能：若外部 AI 能带来新增订单，而交易履约仍掌握在自身手中，平台并不排斥适度开放。
一度被业内调侃为大开环的小红书，便在培育平台电商生态的过程中屡次放开并关停外链跳转。此前小红书曾推出“种草直达”功能，允许广告笔记直接跳转淘宝、京东等外部电商平台页面。这种通过让出前端部分决策流量、换取全网订单与转化收益的做法，正是传统流量平台与履约平台博弈后的妥协产物。
京东长期积累的自营供应链与物流履约体系，是其最坚固的防线。即便用户通过外部AI比价，最终决定成交的，依然是价格、库存与配送时效。外部AI负责激发与筛选需求，京东承接履约，逻辑依然通畅。
当然，京东同样在构筑自己的 AI 防线。用户侧，京东升级了AI购物助手“东东”，并推出JoyAI实时视频数字人，打通购物与生活服务；商家方面，上线京麦AI经营中心，计划在双11期间为超 9 万商家提供数字人直播服务。
京东搞AI的逻辑足够纯粹，即聚焦服务，全面嵌入现有的供应链与交易流程，降低买卖双方的摩擦。
相比京东的高调展示，拼多多的做法在务实基础上，更具隐蔽性。
今年5月，拼多多悄然上线AI搜索功能。用户输入“找低价、限定品牌、黑标或百亿补贴”等复杂长尾需求时，AI会精准拆解条件并匹配商品库。在此之前，平台还陆续上线了试衣间、AI好友互动、AI互动剧等或是促活，或是辅助决策转化的C端功能。
可以看到，拼多多将AI纯粹定位为高效率的内部工具。商品分发也好，促活留存也罢，一系列小功能的上线都带有明确目的性，同时并不会过多干扰用户惯常的购物习惯。这与业内另起炉灶做对话式“导购”，试图改变用户购物习惯的做法形成鲜明对比。
有传言称，拼多多内部至今仍“遵循”古法编程，原因在于，集团并不在意员工是否使用Vibe Coding提效，更在意实际的产出。这种极致的效率导向，从现任拼多多联席董事长赵佳臻早年搭建农产品上行供应链时如出一辙。
商家侧，拼多多官方开放平台长期向商家开放订单与物流接口，允许第三方服务商在授权下开发AI客服与商品管理工具。一位拼多多商家表示，平台小二甚至鼓励使用字节系的电商素材生成工具，还会手把手教学。
京东与拼多多的务实路径表明，AI可以不必急于颠覆消费者的交互入口，而是先从具体、耗时的交易环节切入。
电商是一个极度结果导向的行业。AI帮商家生成一万张营销图，若无法提升转化率，商业价值便大打折扣；AI 替消费者比价十轮，若最终体验不佳，也难以形成复购。
帮商家管生意，替用户做决定。这届双11，电商AI的竞争迁移到了“代理层”。
但无论前端入口如何演变，商品的供给能力、履约时效与售后保障，依然是决定交易能否成立的基石。它能改变流量的流动方向，却无法颠覆最基本的供需规律。
能更高效促成交易的AI才是好AI。
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。
Qin's money.
This year, double 11, there was a special participant in the shopping cart.
It can screen consumers for commodities, prices, coupons and even proactively apply for price refunds; it can also target traders, analyse placement effects, and generate bulk marketing asset. The complex and cumbersome processes that used to be repeated by buyers and sellers are now being taken over by AI.
On 13 October, Ali launched a personal intelligence assistant, "Mo!" to integrate purchasing, price and life services into the AI competency circle. The following day, thousands of questions were asked about the publish version of the office electrician, who tried to manage the cross-platform business with an AI desk.
At almost the same time, Doubao mobile phone assistants met with a collective boycott.
On 14 October, the Doubao mobile phone assistant released an update on the Action Statement on Screen Automation (SAEP), with over 150 applications expressly refusing to be automated by AI. There are many high-frequency applications, such as poaching, payment, beauty groups, and so on.
On the one hand, electrician platforms struggled to embed AI in the shopping chain, and on the other hand, mainstream App was determined to guard against external A interfering with its own trading processes.
AI is reshaping the electrician, but companies are trying to occupy a very different fort.
For the purposes of this question, the electrician is the best access point to the C-PAI Assistant and the B-Enterprise-Agent; for Doubao, shopping is the core scene to validate the ability to perform a cross-application mission; and for Kyoto and Mankind, the value of AI has always revolved around traditional electrician indicators such as selection, conversion, and operational efficiency.
Behind similar capabilities lies a very different business agenda. This new game around the shopping cart is completely surfaced during the double 11th period.
Two books for Ali.
Ali's double 11-year-old version of AI runs in parallel with the user and business.
On the user side, the role of the "shopping consultant" is being upgraded from "shopping agent".
In May this year, thousands of people asked App to be fully connected to the treasure hunt, to access more than 4 billion products, and to complete the discovery, comparison, billing and sale process in natural languages. In addition, there are a couple of questions from the inside of the Zoott App.
The business of poaching and tasting cats is more focused on shelfing, and users need to simplify demand to keyword search, discover and compare the commodities of the mind. Today, in an attempt to recreate interactive logic, such as the purchase of small household furniture, users need only describe living space and budget to enable AI to automatically match and screen.
The 13 October test, “Gorgo!”, went a step further and started to intervene in specific matters, such as the purchase price guarantee, airline settlement and vehicle maintenance request. From “helping consumers find goods” to “doing business for consumers”, the question began to shift from an intelligent search box to a first entry point for users to express their consumption needs.
This is a double 11, which logically becomes a new “shopping agent” examination, with the indicator being the conversion rate of orders, the unit price and the repurchase rate. The two-problem rule is cumbersome and AI calculates best for users, which directly reduces the cost of calculating prices and saves orders that have been abandoned due to cumbersomeness.
On the other hand, the already validated shopping festival is the peak of traffic on the Chinese Internet. High-frequency (HF) electrician transactions are also ideal for the development of super-AI portals, and the test indicator is the user ' s retention habit of referring day-to-day affairs to AI.
Ali's bottom line is ecological closure. A complete compliance system is made up of treasure hunts, payment treasures, Gords and flying pigs. Quaker made a deal, and Quaker turned it into an all-powerful AI portal.
At the commercial level, the office electrician version attempts to replicate this logic to the business service sector.
publish, dated 14 October, provides cross-platform operating boards, trend insights, asset generation and cross-platform ROI optimization, and access to business tools such as business staff, kowtows, and Zoo. publish data show that the electrician industry has more than 20 per cent of its paid users.
Multi-platform operations are normal for current traders, but order, inventory and advertising data are scattered in different backstages. The office tried to become an independent third-party electrician, Agent, bringing scattered data together in the same AI desk.
The ambition of this positioning is that if merchants are accustomed to handing over the Platform ' s statements, marketing programmes and inventory strategies to a question-and-seek office, its mosaics of running context and workflow will turn into an extremely solid B-end software barrier.
Although the core capacity of the Office at this stage is still limited to data integration and decision-making support, the specific placement is still limited to APIpermission and human confirmation, but its strategic intent is well established.
Ari not only uses AI to sell more goods, but also borrows from the power business scene to get questions rooted in the daily workflow of consumers and businesses.
02 Doubao hit whose wall?
If the ability to shop is built on Ali ' s own ecology, then what Doubao cell phone assistants are trying to solve is another ultimate challenge.
Can the user skip App and direct the assignment to AI?
In September this year, based on the automated technology of the mobile phone system permission, the API interface and the GUI (a graphical user interface), the second generation of Doubao mobile phone assistants in Nubian began to try to understand user instructions, simulate manual clicks, slides and input, and cross-applications to perform search, price and order.
This GUI automation model significantly lowers the threshold for accessing the existing App and pushes Agent to the forefront of ecological confrontations.
Simulation of manual operations, while avoiding cumbersome steps for users, is highly accessible to the Platform ' s account permission, payment security and wind-controlled red lines. Deeper conflicts lie in the redistribution of commercial entry points.
The business model of the traditional electrician is based on a “search-recommended-browsing-advertisement” chain, where the platform is realized through exposure and distribution of traffic. Once consumers have entrusted their demand to external AI, the selection and comparison of commodities will be completed outside the App.
When external AI rearranges commodities and distributes flows, the original advertising realization mechanism and recommended algorithms of the electric power platform are at risk of becoming ineffective. The platform may still provide payment and compliance, but loses its core user attention and traffic control. This is why Doubao mobile phone assistants have been subjected to collective resistance.
As mentioned earlier, as of 14 October, the list of rejections of the DoubaoSAEP agreement had been increased to more than 150 pieces, including giants such as peddlers, payment of treasures, beauty packs, a lot of effort, and WeChat.
What needs to be made clear is that the platform resists the automated overstepping of its own application by external assistants, rather than the technology itself. They have their own AI assistants, they have a lot of in-house AI searches, and they welcome AI to enhance the efficiency of their own platforms, but they are determined to prevent external Agent from encroaching on user relationships.
For Doubao, shopping is a litmus test for its ability to cross-apply Agent allocation; but for trading platforms, once scaled up, App will become a purely backstage performance tool. The more AI wants to become an independent super-entry point, the more ecological resistance it is.
It also forces the industry to think in another direction: would it be much less resistance if it did not compete for an independent entrance and only allowed AI to trade with existing electricians?
03 Kyodong and many other transactions
In the rejection list of Doubao, Kyoto is a delicate exception.
As at 14 October, the application of the Kyoto system had not been explicitly included in the rejection list. In business logic, this implies the possibility of a choice: If external AI can bring in additional orders, and transaction performance remains in its own hands, the platform does not exclude moderate openness.
REDnote, once adapted to the industry, was repeatedly released and shut off during the development of the platform electrician ecology. Previously, REDnote had introduced the "Back to the Grass" function, which allowed advertising notes to jump directly to the outside electric power platform pages, such as Treasure and Kyoto. This approach, by giving up the front-end part of the decision-making flow in exchange for a web-wide order and conversion gain, is the result of a compromise between traditional flow platforms and compliance platforms.
The long-established self-employed supply chain and logistics performance system in Kyoto is its strongest defence. Even if the user were to pass the external AI, the final decision would be the price, inventory and distribution time limit. External AI is responsible for stimulating and filtering demand, and Kyoto has a clear logic of compliance.
Of course, Kyoto is also building its own AI line. On the side of the user, Kyoto upgraded the Ai shopping assistant “Toto” and launched the Joy AI live video digital human to get through shopping and living services; on the business side, going online to the Kyomei Operations Centre, it is planned to provide live digital human services to over 90,000 businesses during the double 11th period.
The logic of conducting AI in Kyoto is pure enough, i.e. focused services, fully embedded in existing supply chains and trading processes, and reducing friction between buyers and sellers.
The multiplicity of approaches is more subtle on a pragmatic basis than the high profile of Kyoto.
In May of this year, many of the online AI search features went online. When users enter complex long-term demand such as “low price, branding, black mark, or multi-billion subsidy”, AI will finely dismantle the conditions and match the stores. Prior to that, the platform had been connected to the trial room, to the interaction of AAI friends, to the AAI interactive theatre, or to the C-end function of facilitating decision-making.
As can be seen, much of the effort has been made to position AI purely as an efficient internal tool. The distribution of goods is good, as well as the promotion of survival, and a range of small functions are online with a clear purpose, without excessive interference in the usual shopping habits of users. This contrasts sharply with a dialogue-style “shopping” within the industry, which seeks to change user shopping habits.
There are rumours that there is still “following” the old-fashioned programming in many internal ways, because the group is not concerned about whether employees use Vibe Coding for effectiveness and more about actual output. This extremely efficient orientation is the same as in the early years when the current co-chairperson, Zhao Jia, built an uplink supply chain for agricultural products.
On the side of business, multiple official open platforms have long-term opening of order and logistics interfaces to businesses, allowing third-party service providers to develop AI customer service and commodity management tools under delegated authority. According to a plethora of traders, the platform even encouraged the use of the byte electrician asset generation tool and taught step by step.
Kyodong and his many practical paths suggest that AI can begin by cutting into specific, time-consuming trading chains without rushing to subvert consumer interfaces.
Electricians are an extremely result-oriented industry. The AI business community produces 10,000 marketing maps, which reduce commercial value if conversion rates cannot be increased; AI offers 10 rounds of consumer value, and it is difficult to repurchase them if they end up having poor experience.
Helping with business, making decisions for users. This double 11, the competition of Electrician AI moved to the “agent level”.
However, regardless of the evolution of the front-end entrance, the supply capacity of the commodity, the time limit for performance and post-sale security remain the cornerstones that determine the viability of the transaction. It can change the direction of flows, but it cannot subvert the most basic patterns of supply and demand.
It's good for AI to be able to make the deal more efficient.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.