广告
加载中

京东云:Agent+Coding双轮驱动AI开发走向自动化

龚作仁 2025/11/21 11:51
龚作仁 2025/11/21 11:51

邦小白快读

EN
全文速览

京东云推出Agent+Coding的AI开发新范式,帮助企业实现自动化、智能化的深度应用开发,提高AI生产力。

1.新范式包括JoyAgent智能体平台和JoyCode智能编码平台的深度协同,JoyAgent提供AI算法库和智能体能力,JoyCode可调用这些能力生成应用,两者通过AI Store形成闭环迭代,开发者组合智能体与编码逻辑快速构建定制化AI应用,例如在京东内部已打磨出3.7万个生产级智能体,节省开发时间。

2.AI基础设施升级为JoyScale AI算力平台和JoyBuilder模型训推平台,平台支持混合计算调度英伟达及国产芯片,模型训练与推理效率提升400%,成本降低80%,还提供安全可信数据空间确保合规。

3.实操案例覆盖具身智能、医疗、金融、政务领域,如京东数字人JoyStreamer服务超4万家品牌,在联想案例中带动GMV增长59%并降低运营成本80%,在武汉协和医院落地京医千询大模型服务超230万患者,实现秒级批办服务。

京东云技术推动数字技术在品牌营销和产品创新中的应用,通过AI智能体提升品牌互动和消费体验。

1.在品牌营销方面,京东数字人JoyStreamer已服务超4万家品牌,从直播间扩展到多元化场景,帮助打破创意边界,如联想11.11活动GMV同比增长59%,运营成本降80%,数字人IP与产品联动增强用户粘性。

2.消费趋势上,数字代言人如恩雅吉他的Aura和山西文旅的“复活”大佛,提升品牌故事性和用户行为参与,京东云JoyAI大模型支持智能体行为优化,响应市场对个性化互动的需求。

3.产品研发启示包括使用Agent+Coding范式简化开发,品牌可参考京东开源工具如2400+MCP工具和350+设计模板,快速构建定制应用,结合AI基础设施提升效率,例如算力平台降低开发成本和门槛。

京东云的AI工具和平台提供增长机会与应对措施,帮助企业捕捉AI市场红利并规避风险。

1.机会提示集中在消费需求变化,如Agent+Coding新范式降低代码门槛,使业务人员创意快速转化为企业应用,JoyAgent在GAIA评测准确率超77%,GitHub开源收获超1万Star,卖家可学习其模型快速落地深度应用。

2.最新商业模式强调平台合作方式,京东云全面开源AI算法库、2400+MCP工具等,支持合作伙伴免费使用,并联合华为昇腾等企业强化混合计算,卖家可通过合作接入AI Store优化运营。

3.风险规避和应对包括安全合规方面,京东云数据空间100%采用国密算法实现全链路加密“可用不可见”沙箱,确保数据流通;同时,JoyBuilder平台通过云原生系统秒级扩缩容,保障服务稳定性,成本降80%,减少技术投入风险。

京东云的具身智能解决方案为工厂提供生产优化启示,推动数字化和产品创新。

1.产品生产设计需求启示于具身智能应用,京东开放20多个核心业务场景如物流、健康,帮助企业优化算法,使用JoyAgent智能体提升设备对话轮次平均超120%,未来5年计划采购300万台机器人用于物流,实现自动化升级。

2.推进数字化的商业机会体现在AI开发工具使用,工厂可借鉴JoyCode智能编码平台高效解决代码维护难题,结合开源工具如AI算法库快速定制模型;京东云与机器人和AI玩具企业技术对接的JoyInside平台,提供从算力到实战工具的全套方案。

3.电商启示源于真实数据和成本降低,如京东云平台提升模型效率400%、成本降80%,工厂可学习混合计算调度英伟达及国产芯片的实践,推进自身设施数字化。

行业趋势聚焦AI开发自动化和新技术应用,京东云工具解决客户痛点。

1.行业发展趋势呈现Agent+AI主导深度应用开发,新范式JoyAgent与JoyCode协同通过AI Store实现闭环,加速向全业务流程渗透;全球评测中JoyAgent准确率超77%位列第一梯队,体现复杂任务通用性。

2.新技术包括开源多模态RAG技术和JoyCode-Agent以74.6%通过率解决大型代码库问题,京东云还推出JoyScale算力平台支持GPU中心混合计算,提升效率400%;数据安全方面采用国密算法打造可靠数据空间。

3.客户痛点和解决方案围绕开发门槛高与资源需求,JoyAgent与JoyCode无缝集成让非代码人员创意快速落地,如企业级应用优化;工具提供500种算法工具链和可扩展云平台,助服务商高效定制模型并规避安全风险。

京东云平台的最新做法强化运营管理和合作模式,应对商业需求。

1.平台的最新做法包括开源开放战略,全面向伙伴开放AI算法库和开发组件,JoyScale算力平台整合华为昇腾、海光等国产芯片,内核统一调度实现高效自主算力管理,并通过高可用云原生系统秒级扩缩容优化运营。

2.平台招商策略以联合为主,与40余家机器人和AI企业技术对接提供工具支持;安全风控强调数据合规,100%国密算法保障“可用不可见”,规避监管风险;运营管理提升体现在模型训练成本降80%。

3.商业对平台的需求回应聚焦AI基础设施缺口,京东云JoyBuilder平台集成50种开源模型,提供算法工具链,助平台商简化模型构建;案例如金融增长云服务国有四大行3家,显示平台在政务、医疗中的高效批办能力。

产业新动向以Agent+AI范式和数据合规为主导,京东云模式提供政策启示。

1.产业新动向表现为AI开发从单点转向深度应用,新范式Agent+Coding形成闭环迭代,推动AI生产力重塑;商业模式创新体现在京东开源的商业价值验证,如JoyAgent开源收获超1万GitHub Star, JoyCode助力维护大型复杂代码库。

2.新问题如基础设施缺口和合规风险,京东云倡导AI Infra 1.0标准化,JoyScale平台支持GPU中心混合计算;数据安全作为关键问题,建议全链路加密“可用不可见”沙箱技术保障流通,可延伸至政策法规应用。

3.政策建议启示源于真实案例,如京小亦政务助手实现1500项事务秒批秒办,凸显AI助政务高效;研究可分析京东医疗京医千询模型在多医院落地,数据驱动诊疗效率提升230万服务人次,为产业标准提供参考。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

JD Cloud has launched a new AI development paradigm, Agent+Coding, designed to help enterprises achieve automated and intelligent deep application development, thereby boosting AI productivity.

1. This new paradigm involves the deep synergy between the JoyAgent intelligent agent platform and the JoyCode intelligent coding platform. JoyAgent provides AI algorithm libraries and agent capabilities, which JoyCode can call upon to generate applications. The two platforms form a closed-loop iteration via the AI Store, enabling developers to quickly build customized AI applications by combining agents and coding logic. For instance, JD.com has internally refined 37,000 production-level agents, significantly saving development time.

2. The AI infrastructure has been upgraded to the JoyScale AI computing platform and the JoyBuilder model training and inference platform. The platform supports hybrid computing scheduling for both NVIDIA and domestic chips, increasing model training and inference efficiency by 400% while reducing costs by 80%. It also provides a secure and trusted data space to ensure compliance.

3. Practical applications span embodied intelligence, healthcare, finance, and government affairs. For example, JD's digital human, JoyStreamer, serves over 40,000 brands; in a Lenovo case study, it drove a 59% GMV increase and reduced operational costs by 80%. At Wuhan Union Hospital, the Jingyi Qianxun large model serves over 2.3 million patients, enabling second-level approval services.

JD Cloud's technology promotes the application of digital tech in brand marketing and product innovation, enhancing brand interaction and consumer experience through AI agents.

1. In brand marketing, JD's digital human JoyStreamer has served over 40,000 brands, expanding from live streams to diverse scenarios to break creative boundaries. For example, during Lenovo's 11.11 campaign, GMV grew 59% year-over-year with an 80% reduction in operational costs, and the digital human IP's linkage with products strengthened user engagement.

2. Regarding consumer trends, digital spokespersons, like the Aura for Enya Guitar and the 'revived' Buddha for Shanxi tourism, enhance brand storytelling and user participation. JD Cloud's JoyAI large model supports agent behavior optimization, responding to the market's demand for personalized interaction.

3. Insights for product R&D include using the Agent+Coding paradigm to simplify development. Brands can leverage JD's open-source tools, such as 2400+ MCP tools and 350+ design templates, to quickly build custom applications, combined with AI infrastructure to improve efficiency, e.g., the computing platform lowers development costs and barriers.

JD Cloud's AI tools and platforms present growth opportunities and countermeasures, helping businesses capture AI market dividends and mitigate risks.

1. Opportunities focus on shifting consumer demands; the Agent+Coding paradigm lowers the coding barrier, allowing business personnel to quickly transform ideas into enterprise applications. JoyAgent achieved over 77% accuracy in the GAIA benchmark, and its GitHub open-source project garnered over 10,000 Stars. Sellers can learn from its models for rapid deployment of deep applications.

2. The latest business models emphasize platform partnerships. JD Cloud has fully open-sourced its AI algorithm library, 2400+ MCP tools, etc., supporting partners' free use, and collaborates with firms like Huawei Ascend to strengthen hybrid computing. Sellers can optimize operations by partnering to access the AI Store.

3. Risk mitigation includes security and compliance. JD Cloud's data space uses 100% domestic cryptographic algorithms for full-link encryption and a 'usable but invisible' sandbox, ensuring secure data flow. Meanwhile, the JoyBuilder platform enables second-level scaling via a cloud-native system, ensuring service stability with an 80% cost reduction, minimizing technical investment risks.

JD Cloud's embodied intelligence solutions offer insights for production optimization in factories, driving digitalization and product innovation.

1. Insights for product production and design stem from embodied intelligence applications. JD has opened over 20 core business scenarios, like logistics and health, helping enterprises optimize algorithms. Using JoyAgent, the average dialogue turns for equipment interaction exceed 120. JD plans to deploy 3 million robots in logistics over the next five years for automation upgrades.

2. Commercial opportunities for advancing digitalization lie in using AI development tools. Factories can leverage the JoyCode intelligent coding platform to efficiently solve code maintenance challenges and use open-source tools like the AI algorithm library for rapid model customization. The JoyInside platform, which connects with robot and AI toy companies, provides a full suite of solutions from computing power to practical tools.

3. E-commerce insights derive from real data and cost reductions. For instance, JD Cloud's platform improves model efficiency by 400% and cuts costs by 80%. Factories can learn from the practice of hybrid computing scheduling for NVIDIA and domestic chips to advance their own facility digitalization.

Industry trends focus on AI development automation and new technology applications, with JD Cloud's tools addressing client pain points.

1. The industry trend shows Agent+AI leading deep application development. The new JoyAgent and JoyCode paradigm, synergizing via the AI Store, creates a closed loop, accelerating penetration into full business processes. In global evaluations, JoyAgent's accuracy exceeds 77%, placing it in the top tier, demonstrating versatility in complex tasks.

2. New technologies include open-source multimodal RAG and JoyCode-Agent, which solves large codebase issues with a 74.6% pass rate. JD Cloud also launched the JoyScale computing platform supporting GPU-centric hybrid computing, boosting efficiency by 400%. For data security, it employs domestic cryptographic algorithms to create a reliable data space.

3. Client pain points and solutions revolve around high development barriers and resource demands. The seamless integration of JoyAgent and JoyCode allows non-coders to quickly implement ideas, such as enterprise application optimization. The tools offer a 500-algorithm toolchain and a scalable cloud platform, helping service providers efficiently customize models and mitigate security risks.

JD Cloud's latest platform strategies enhance operational management and partnership models to address commercial demands.

1. The platform's latest approach includes an open-source strategy, fully opening AI algorithm libraries and development components to partners. The JoyScale computing platform integrates domestic chips like Huawei Ascend and Haiguang, enabling unified kernel scheduling for efficient, autonomous computing management. High-availability cloud-native systems allow second-level scaling to optimize operations.

2. The platform's merchant acquisition strategy emphasizes collaboration, providing tool support through technical integration with over 40 robot and AI companies. Security and risk control focus on data compliance, using 100% domestic cryptographic algorithms for 'usable but invisible' protection to avoid regulatory risks. Operational improvements are evident in an 80% reduction in model training costs.

3. The platform's response to commercial needs addresses the AI infrastructure gap. The JoyBuilder platform integrates 50 open-source models and provides an algorithm toolchain, helping marketplace sellers simplify model building. Case studies, such as serving three of the four major state-owned banks with financial growth cloud services, demonstrate the platform's efficient batch-processing capabilities in government and healthcare.

Industry trends are dominated by the Agent+AI paradigm and data compliance, with JD Cloud's model offering policy insights.

1. The industry trend shows AI development shifting from point solutions to deep applications. The new Agent+Coding paradigm forms a closed-loop iteration, driving a reshaping of AI productivity. Business model innovation is evidenced by the commercial validation of JD's open-source projects; for example, JoyAgent's open-source garnered over 10,000 GitHub Stars, and JoyCode aids in maintaining large, complex codebases.

2. New challenges include infrastructure gaps and compliance risks. JD Cloud advocates for AI Infra 1.0 standardization, with the JoyScale platform supporting GPU-centric hybrid computing. Data security is a key issue; it recommends full-link encryption and 'usable but invisible' sandbox technology to safeguard data flow, which can be extended to policy and regulatory applications.

3. Policy recommendations are informed by real-world cases. For instance, the Jing Xiao Yi government assistant handles 1,500 tasks with instant approval, highlighting AI's efficiency in public services. Research can analyze the deployment of JD's medical Jingyi Qianxun model across multiple hospitals, which data-drivenly improved diagnostic efficiency, serving 2.3 million patients, providing a reference for industry standards.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

深度应用为核,重塑AI生产力。11月20日,京东云城市大会在南京举行。会上,京东云展示了JoyAI大模型落地行业的领先技术和应用实践,持续升级JoyAgent智能体等五大平台,以及覆盖具身智能、医疗、金融、政务的四大行业解决方案,以Agent+ Coding双轮驱动的AI开发新范式,助力企业加速落地深度应用,重塑AI生产力。

京东集团技术委员会主席、京东云事业部总裁曹鹏表示,“Agent+ Coding的AI开发新范式,正在成为AI应用开发的主流,这种以Agent简化代码、代码反哺Agent的闭环,将让AI开发越来越自动化、智能化,最终实现‘从开发者少写代码,到开发少写代码’。同时,大规模应用爆发,也将推动AI基础设施迈向标准化,AI Infra 1.0已开启。”

Agent+Coding深度协同,构建AI开发新范式

虽然“超级应用”还有距离,但聚焦企业端的“深度应用”已奔涌而至,AI应用正在从单点应用,转向全业务流程渗透,一种新的AI开发范式也随之涌现。

京东云将JoyAgent智能体平台与JoyCode智能编码平台紧密协同,构建支持深度应用的AI开发新范式。JoyAgent所提供的智能体、AI算法库等原子能力,可直接被JoyCode调用和集成;JoyCode生成的应用可反馈至JoyAgent,用于优化智能体行为与知识模型,两者通过AI Store实现无缝协同,形成持续进化的闭环,开发者可灵活组合智能体能力与编码逻辑,快速构建复杂、深度定制化的AI应用。

京东云能够将智能体平台与代码平台深度融合,源于三大优势:

技术栈同源同构。JoyAgent和JoyCode两个产品都成长于京东云团队,两者基于同一技术栈,在编码逻辑、通信协议、开发组件、算法库等方面同源同构,可实现更紧密协同。

坚持开源开放。京东云将成熟组件全面向合作伙伴开放,包括全套开箱即用的AI算法库,2400+MCP工具、350+设计版式模版等,并最新开源多模态RAG技术,让开发者们能够快速获得与京东云一致的技术能力。

生产级智能体。历经京东内部大规模业务场景锤炼,JoyAgent产品可靠性得到验证,京东已打磨出3.7万个稳定可靠的生产级智能体,充分验证了Agent+ Coding的商业价值。

JoyAgent与JoyCode的领先性已经获得行业认可。

JoyAgent在GAIA智能体评测中,验证集、测试集准确率分别超77%和67%,超越多个知名团队,位列全球第一梯队,展现复杂任务中的通用性优势,自2025年7月开源以来,JoyAgent在GitHub上已收获超1万Star。

JoyCode通过规约编程端到端智能体与CSR上下文引擎,高效解决大型复杂代码库维护难题。在SWE-Bench Verified基准测试中,JoyCode-Agent以74.6%通过率位居全球前三,体现卓越的编程问题解决能力,并已在GitHub开源。

过往,企业中有大量业务人员拥有宝贵的场景洞察,却受限于代码开发门槛。现在,基于JoyAgent与JoyCode的深度融合,每一个创意都可迅速转化为企业级深度应用。

基于JoyAI大模型,京东数字人JoyStreamer已服务超4万家品牌,从直播间走向多元化应用,带领数字人从替身执行走向个性创造。以联想为例,今年11.11,京东数字人带动直播间GMV同比增长59%,运营成本降低80%,京东数字人IP还与联想拯救者系列产品实现跨界联动。从恩雅吉他的数字代言人Aura,到山西文旅的“复活”大佛,数字人技术正帮助各行业打破创意边界,实现增长。

AI Infra1.0驱动基础设施规模化部署

随着大模型应用的深入,以CPU为中心的架构在支持AI原生应用上面临挑战,需要以GPU为中心重塑基础设施;此外,面对激增的推理需求,计算资源持续增加,需要以混合计算平台为基础,构建一套以GPU为核心的AI Infra。

在混合计算方面,京东在行业里最早大规模部署混合算力,拥有超3700万核容器集群,历经618、双11、春晚等极限场景检验,建立起行业最领先的混合算力技术储备,这些优势在京东云JoyScale AI算力平台中充分体现,平台不仅跟华为昇腾、海光、寒武纪等国产芯片深度合作,而且实现从内核层统一调度英伟达及国产芯片,为企业提供高效自主的算力管理。

在模型训推方面,京东云JoyBuilder打造了三大核心优势:首先,得益于自研的高性能推理引擎,模型训练与推理效率提升400%;其次,平台集成50种开源模型和数据集,提供100余种算法工具链,让企业能基于自身数据快速构建专属大模型;最终,通过高可用云原生系统实现的秒级扩缩容能力,在保障服务稳定性的同时,总体应用成本降低80%。

除了模型和算力,数据作为人工智能三要素之一,数据的安全合规是决定AI应用效果的关键。京东云基于多年数据安全技术沉淀,打造了安全可信数据空间,100%采用国密算法,实现全链路加密防护,通过安全可信加工能力,在"可用不可见"的安全沙箱中使用,确保数据流通的安全可靠。

全新升级四大行业解决方案,重塑AI生产力

技术价值最终体现在真实场景的落地中。京东云发布面向具身智能、金融、医疗健康、政务等多行业的全栈解决方案,并与伙伴共享真实数据集,共同打磨AI在复杂环境中的实战能力。

在具身智能方面,京东已开放物流、健康、零售等20多个核心业务场景,帮助企业在真实环境中优化算法。在平台共建方面,京东JoyInside平台已实现与40余家头部机器人和AI玩具企业的技术对接,设备对话轮次平均提升超120%;在规模化部署方面,京东计划未来五年在物流领域采购300万台机器人、100万台无人车和10万架无人机,以实际订单推进产业链成熟。除了支持具身智能企业产品落地,京东自身也在探索边缘近场算力和世界模型等具身智能领域的新技术。

基于这些积累,京东云正式推出具身智能行业解决方案,通过提供从万卡算力到实战工具的全套平台,服务具身智能企业从模型训练与具身Agent开发的全流程。同时,依托京东自研的JoyInside附身智能,为智能体注入环境感知、语音交互与动作规划的核心能力。

在医疗方面,在医疗方面,京东京医千询大模型是⾏业⾸个可信推理 、全模态医疗⼤模型,全⾯服务复杂专病诊疗。基于其打造的京东卓医正在医院全场景应用。目前,已在武汉协和医院、苏州市立医院等多家头部三甲医院落地,服务患者超230万人次。京东云推出深度融合京医千询大模型的医疗解决方案,它以京东云智算为底座,依托京医千询医疗大模型,实现模型与应用无缝衔接。

在金融方面,结合京东金融APP超过10年的实践经验的金融增长云,过去几年为金融机构持续增长提供战术咨询诊断、营销平台建设、联合运营服务以及数字基础设施等,国有四大行中,3家使用了京东云的金融云。融合大模型的新一代金融增长云也在进化中,不仅有面向过去场景升级的大模型应用产品,如营销、数据Agent等 ,还有基于京东金融场景实践的完全AI Native的产品,京东金融正在大面积使用。

在政务方面,依托京办平台,京东云推出京办智能政务解决方案,深度融合京东大模型与JoyAgent能力,为政府公务人员提供智能会议、文档与数据问答等政务智能体服务。以北京经开区“京小亦”政务助手为例,它已实现超1500项事务的智能咨询,并在居住证办理、人力资源等场景中实现“秒批秒办”,让政务服务更高效、更便捷。

大模型浪潮汹涌澎湃,京东云将持续深耕技术,以更开放、更协同的生态理念为引领,不断拓展和深化大模型技术的边界与应用。同时,凭借在京东内部场景千锤百炼的深度应用经验,打造出的安全稳定、开放、极致性能、极致性价比的技术与产品,助力更多企业重塑AI生产力。

注:文/龚作仁,文章来源:Laborer,本文为作者独立观点,不代表亿邦动力立场。

文章来源:Laborer

广告
微信
朋友圈

这么好看,分享一下?

朋友圈 分享

APP内打开

+1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0