最近在开源社区GitHub上，一款AI专利附图工具开源上线了，出人意料的是，该AI产品来自一家做了多年充电宝、充电器的硬件公司——品胜。
在这一专利附图工具开发出来后，品胜创始人赵国成原本考虑过商业化，但旗下AI产品的负责人丛泽龙却认为，与其先考虑怎么收费，不如把工具开放出去，让更多企业能在自己的环境里使用，也让更多开发者参与测试和迭代。赵国成最终接受了这个建议。
这不是品胜唯一要开源的AI产品。作为集团AI产品负责人的丛泽龙告诉我们，团队还开发完成了一套用于筛选AI Native人才的线上“AI试炼”工具，也将计划开源。与此同时，品胜正在制作AI Lab官网，准备把AI管理提效工具、Agent产品下载入口和使用指南集中起来一起推出。
在丛泽龙的设想里，品胜未来的AI创新工具都将走向开源。但他没有把这条路描述成一场已经胜利的战役：工具做出来只是开始，后面有没有人持续使用，其他公司能不能用得起来，是否真的解决了问题，还需要继续打磨。
这让品胜的AI实践有了一个不同于常见企业转型故事的起点。它不仅关乎一家传统制造企业如何用AI提效，也关乎企业内部产生的创新成果，究竟应该被包装成一款收费产品，还是被放进更大的开源生态里。
打开一道AI创新的口子
别看丛泽龙是00后，已经在品胜带起了一支年轻的AI产品开发团队，他们的成长故事可以从一张专利附图说起。
在一次产品研发过程中，专利工程师与结构工程师之间的沟通效率不高，相关工作拖慢了进度。问题抛给了丛泽龙团队。
专利附图看起来只是研发链条中的一环，实际却涉及产品结构、工程模型和专利规范。结构工程师手里的3D模型，并不能直接变成符合要求的专利图纸。过去，相关人员需要反复沟通模型细节、视图选择和部件标注，工作推进往往受制于跨岗位协作。
丛泽龙团队尝试把AI与开源CAD工具结合起来，让系统读取3D模型、理解产品结构，并辅助生成爆炸图、部件标注等专利附图。AI负责理解人的意图、识别模型中的结构，再调用工程软件完成具体处理。它不是凭空生成一张“看起来正确”的图片，而是围绕真实模型开展工作。
这一区别很重要。专利图纸必须对应产品实际结构，不能只追求视觉上的完整。AI可以减少前期处理和反复沟通，但生成结果仍需要专业人员检查，最终质量也不能只靠模型自己判断。
据团队的实践设想，原本需要经历较长沟通周期的部分前期工作，有机会缩短到几十秒或几分钟。不过，这并不意味着从产品模型到正式专利申请的整个流程都能被压缩到几分钟，工程校验和专业审核仍然不可缺少。
工具做出初步成果后，团队才向老板赵国成展示。一个产品要不要做，并不始于公司层面先组织一场AI战略会议，这恰恰是品胜正在进行的AI转型的特点：创新不是从一份自上而下的规划文件开始，而是从业务流程中一个具体、令人头疼的问题开始。
丛泽龙说，最初的尝试并没有先变成一个正式立项的项目，而是先在部门内部找到需求、动手做，再把完整流程拿出来讨论。
这种做法给了团队一定的试错空间。与其先证明一个AI项目值得投入，再申请资源，不如先从真实工作中找到问题，做出可运行的东西，再观察它是否有价值。
工具逐渐成形后，赵国成看到了它的商业潜力。按照传统软件的思路，这样一款能解决实际问题的工具，似乎可以收费。但丛泽龙提出了另一种考虑：专利涉及企业尚未公开的研发信息，如果使用工具必须上传数据，潜在用户可能会有所顾虑；开源并支持本地部署，反而更有机会降低企业的使用门槛。
一个在制造业摸爬滚打多年的创始人，首先看见产品的商业可能；一个年轻团队负责人，则希望把产品放进更大的生态里。两种判断在一次具体的产品决策中相遇，最终推动品胜选择开源。
赵国成后来这样解释自己的判断：“如果按照传统SaaS软件企业的业务模式，这个产品收费也是行得通的。”但他也知道，品胜一旦下决心扎实地推进AI在企业的落地，类似专利附图工具这样的AI产品创新案例会源源不断，并且AI Native组织也将呼之欲出，“开源才是我们的护城河，未来这个团队和这条路径会更具价值，收费反而自缚手脚。”
当然，产品开源并不自动等于护城河，未来能否带来更多使用者、外部反馈和持续迭代，仍要看产品本身是否真正解决问题。但对品胜而言，这次选择至少意味着，团队的目标不再只是完成一个内部工具，而是尝试把内部积累的经验转化为外部也能使用的能力。
从“做一个工具”到搭建工具箱
专利附图工具之后，品胜还在推进其他AI应用，其中一项来自招聘。
丛泽龙向我们介绍团队开发的“AI试炼”工具，希望帮助公司用它来筛选符合AI Native人才定义的候选人，未来也计划向外部开源。AI Lab官网也正在制作中，准备集中呈现品胜AI产品创新团队开发的工具入口、下载方式和使用指南。
这些尝试有一个共同点：它们不是先确定一个面向市场的产品，再寻找使用场景，而是从品胜内部的具体需求中生长出来。专利附图工具对应研发、法务协作，AI试炼对应人才筛选，其他管理提效工具则试图处理日常工作中的重复劳动。
但内部好用的工具，不一定能直接成为外部产品。不同企业的数据结构、工作流程和安全要求并不相同，开源也不意味着别人下载后就能立刻使用。丛泽龙更关心的是，工具发布后能否获得真实反馈，再根据使用情况继续完善。
这也是品胜准备把相关工具集中起来的原因。若每个工具都散落在不同团队手中，其他员工不知道去哪里找，外部用户也难以了解有哪些产品可用。官网和使用指南试图解决的，是工具开发之后的分发与使用问题。
从这个角度看，品胜要建立的并不只是一个工具清单，而是一条从业务需求出发、经过内部试用、再走向外部验证的路径。
不过，工具数量并不能直接证明转型成效。真正的问题是，工具能否进入日常工作，是否有人愿意持续使用，以及它究竟节省了多少时间、改善了哪些结果。这些问题，品胜仍需要用后续实践回答。
AI正改变管理模式
如果说专利附图工具展示了AI如何解决具体业务问题，那么品胜内部的“许愿池”和“还愿群”，则呈现了另一面：当做工具的门槛下降，企业会涌现更多需求，但如何管理这些需求，同样成为一道新题。
丛泽龙的团队会收集各部门提出的AI需求，评估后再决定如何处理。据悉，“许愿池”中每周约有两三个需求会得到推进，“需求并不总能立即满足，团队的人手有限，也需要判断哪些问题值得优先投入。”
品胜得益于这一机制，把AI应用从少数技术人员的探索，变成了可以被其他部门提出、反馈和参与的工作。员工不必先掌握复杂的技术，才有机会尝试改善流程；他们可以先提出问题，再由团队一起寻找解决办法。
但需求一多，新的问题随即出现了：哪些需求值得做？做出来的工具是否有人用？不同部门的相似需求能不能复用？如果每个问题都重新开发一次，AI反而可能制造更多维护成本。
品胜围绕Skill开展的探索，正好碰到了这一矛盾。
Skill可以理解为让AI重复完成某类任务的一套指令、流程和知识。品胜AI产品团队自去年年底开始关注这一概念，今年3月起，他们陆续为不同部门和同事制作起Skill，“希望把一些工作方法沉淀下来，让员工不必每次从头操作。”丛泽龙在实践中，很快发现一些问题暴露出来。
像制作一个Skill所花的时间，有时甚至超过使用它节省的时间；工具还需要持续维护，使用频率和实际效果也不容易追踪；如果每个员工都单独制作一套，既容易重复，也很难判断哪些值得保留。
于是，丛泽龙团队开始制作SkillHub，希望把分散的Skill集中起来，便于查找、复用和管理。
听起来，这像是一个自然的解决方案，但真正落地并不简单。Skill应该在什么时机被调用？使用者能否判断它的质量？如果效果不好，应该由开发者还是使用者来优化？文档、知识库与Skill之间又该如何组织，才能让积累下来的经验在需要时真正被调用？
这些问题，丛泽龙并没有现成答案。这也揭示了企业AI落地中一个容易被忽视的事实：AI降低了制作工具和整理知识的门槛，却没有自动消除组织协作的成本。一个工具能被快速做出来，不代表它能被长期维护；一份知识被保存下来，也不代表它会在正确的时间被正确的人使用。
对企业而言，AI不只是多了一种软件，更意味着要重新考虑工作流程、知识沉淀和责任分配。工具越容易开发，企业越需要建立判断需求、管理工具和衡量效果的机制。
与丛泽龙从具体工具切入不同，赵国成更关注AI能否改变品胜的核心业务效率。
赵国成虽不是一个高频使用AI工具的人，但他告诉我们，自己每天会花大量时间透过短视频介质，去学习国内外最新的AI研究成果，然后去到丛泽龙团队的办公区域，进行讨论、分享等思想碰撞，研究哪些成果能应用于品胜等企业的实际工作中。
“我关心的重点，是研发环节的AI提效。”在赵国成看来，品胜的产品研发与创新高度依赖工程师经验、数据验证多且关系复杂，如果AI提效成功，将会缩短研发周期，提升研发成功率，这才是真正的生产力提升。
作为硬件企业，品胜开发一款产品，不只是完成设计图或写出一段代码，还要把电子方案、元器件、采购价格、供应风险、生产组装和验证环节连接起来。一个环节出现变化，都可能影响整个产品的成本和进度。
以一款充电器为例，从产品需求文档（PRD）出发，电子工程师需要选择核心部件并评估参数，采购人员要判断价格，供应链还要核查交付周期和停产风险。方案进入组装和验证阶段后，团队还要评估成本是否符合目标。如果成本超出预期，就可能需要调整定价，或者重新设计方案。
这是一条依赖经验、数据和跨部门配合的链条。即使每个环节单独看都不算复杂，把它们串起来，也可能耗费大量时间。
按照品胜团队的描述，相关方案反复沟通、评估和调整，一个月甚至更长的周期并不罕见。
问题在于，很多经验并没有完整沉淀在系统里。工程师知道某个元器件的参数是否合适，也可能记得某家供应商过去出现过什么问题，但这些判断常常分散在个人经验和不同资料中。人员变化、产品切换或工具更换，都可能让原有经验难以复用。
品胜并非从零开始补数据。据了解，自2015年起，品胜开始使用产品生命周期管理系统（PLM），持续积累物料信息、设计规范和风险记录。丛泽龙团队在这些数据基础上搭建元器件数据库，并连接第三方及原厂平台，同步元器件接口、电压协议、规格等信息。
有了这些基础，AI才有机会参与更复杂的研发任务。
目前，丛泽龙团队正在探索完全用 AI 做出一款 Agent 遥控手柄。设想中，用户拿起手柄，圈选电脑屏幕上的内容，通过语音交互。围绕这款产品，团队尝试把需求定义、器件选型、电路与结构设计、固件及上位机开发，以及样机试制、整机验证串成一条连续的AI研发流程。
团队已完成AI器件选型和初步结构堆叠，AI 也已生成原理图和初步 PCB Layout。眼下，团队正结合 Agent 复核与工程师评审，对照器件数据手册，检查接口选型、天线周边布局和布线约束，并根据发现的问题调整设计。
接下来，团队还需推进制板、焊接和上电调试，再开展外设驱动与软硬件联调，通信传输到上位机响应等完整链路，让屏幕里的设计逐步变成可以实际操作的样机。
但这仍是一项需要验证的实践。元器件数据是否完整、供应商信息是否及时更新、模型判断是否可靠，都会影响结果。硬件产品还要经过实际测试和专业评审审核，AI给出的方案不能绕过这些环节。
赵国成将这一方向称为从 “Vibe Coding” 走向 “Vibe Engineering”。他期待，把“提需求，让 AI 写代码”的方式延伸到硬件产品，让 AI 生成方案、操作工程工具、检查并修改设计，人则在关键节点判断、拍板。制造企业积累的经验由此成为指导 AI 工作的依据，缩短从一个想法到一个硬件产品的距离。
制造老炮与年轻团队的分工
品胜的AI探索，最终落在两种不同的工作范式上。
赵国成有多年硬件制造经验，熟悉产品、供应链和企业经营。他关注AI的最新进展，也会追问这些技术能否真正解决企业的问题。
丛泽龙团队则更靠近具体执行：从员工抱怨的流程中发现需求，动手制作工具，再把使用中暴露的问题带回去调整。
两者并非简单的“老板出钱、年轻人做事”。专利附图工具是否收费、是否开源，就是一次围绕产品路线的讨论。赵国成看到了商业化的可能，丛泽龙提出了安全、部署和生态方面的考虑，最终公司选择先开放工具。
赵国成把团队在AI上的尝试看作一种长期投入，在他看来，品胜扎实推进AI落地，在相关创新工具产品层出不穷背后，AI Native组织也可能随之形成。
所谓AI Native，并不只是员工会使用提示词或掌握几款工具。更重要的是，企业能否重新组织工作：把原本依赖个人经验的环节沉淀下来，把跨部门反复沟通的问题变成可复用的流程，让员工能够更快发现问题、试做方案，再通过真实使用来验证。
品胜已有一些条件可以支持这种尝试。它有真实的产品研发流程、积累多年的物料和设计数据，也有供应链及生产环节作为应用场景。团队不是在抽象地讨论AI能做什么，而是能把工具放进实际工作中，观察它是否有效。
其他企业未必没有做类似的事，之所以像专利附图工具这样的产品诞生在品胜，丛泽龙感慨，品胜把AI创者的优先级和投入方向，从一号位视角做了明确。
据悉，在赵国成授意后，不论品胜公司的整体AI化转型还是有关AI的分享、评审、招聘和考核，都由丛泽龙部门主导。丛泽龙也讲述，在品胜内部组织的AI宣讲会、创新活动中，公司会给所有分享的同事派发AI应用奖金。当前，品胜不仅愿意在AI创新的方向上多做尝试，还把其中一些成果开源，让更多人参与检验。
这条路并非没有风险。在交流中，丛泽龙也顾虑一些内部工具只适用于品胜自己的流程；一些部门同事提出的需求未必都值得产品化；甚至已经开源的项目，如果缺少使用者和维护力量，同样可能逐渐停滞。即使AI试炼、SkillHub和专利附图工具都已形成初步成果，也不能据此认定企业已经完成了AI转型。
品胜接下来仍要面对一些更具体的问题：AI工具是否被持续使用，外部企业能否顺利部署，研发流程有没有可衡量的改善，节省下来的时间能否转化为更短的产品周期和更高的研发成功率。
对赵国成而言，AI最终要落到生产力上；而考验丛泽龙团队的，将是怎么把一次次具体的创新变成可以复用、持续更新的能力。
眼下，品胜的AI变形记，还远没有到可以下结论的时候。它的意义或许不在于已经做出了多少款工具，而在于一家传统硬件企业开始尝试改变解决问题的方式了：让创新不必总是等待正式立项，让年轻团队能够从业务现场出发，也让内部做出来的东西接受更广泛的使用和检验。
从一张专利附图开始，品胜试图搭建的，是一条连接真实需求、工具开发、组织协作与外部验证的路径。它能否成为新的竞争力，最终仍要由使用效果来回答。（本文首发于钛媒体APP，文｜钱方）
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Recently, in GitHub, an AI patent-based map-based tool in the open-source community, unexpectedly from a hardware company that has been charging for many years - the Winners.
After this patented map-based tool was developed, Zhao Guo-Sheng, the founder of the success story, originally considered commercialization, but the chief of the flagship AI product, Jun Zelong, argued that instead of first considering how to charge, it would be better to open up the tool so that more companies could use it in their own environment, and to involve more developers in testing and iterative processes. Zhao Zheng Cheng finally accepted the recommendation.
It's not the only AI product that has to open up. As the group leader of AI's products, Szelong told us that the team has also developed an online “AI test” tool for screening AI Native talent, which will also open up the project. At the same time, it is developing the AI Lab Web of Officials, which will bring together the AI Management Efficiency Tool, Agent ' s download portal and user guide.
In the Mong-Zelong vision, the AI innovation tool for the future of winning will move to the open source. But he did not describe the road as a battle that had been won: the tools were just the beginning, whether they were being used on a continuous basis, whether other companies could use them, whether they really solved the problem and whether they needed to continue grinding.
This gives a starting point for successful AI practices that are different from common business transformation stories. It is about not only how a traditional manufacturing firm uses AI to improve its performance, but also about the innovations that are generated within the enterprise, whether they should be packaged as a fee-paying product or placed in a larger open-source ecology.
Open an AI innovation mouth.
Don't look at the Zelong after 0000, they've brought a young AI product development team to the competition, and their growth stories can start with a patented map.
During a product development process, communication between patent engineers and structural engineers was inefficient and the related work slowed down. The problem is left to the team.
Patent attachments appear to be only a link in the R & D chain, but actually involve product structures, engineering models and patent specifications. The 3D model in the hands of structural engineers does not directly translate into a patented drawing that meets the requirements. In the past, the personnel involved had had to communicate repeatedly with the details of the model, the view selection and the labelling of components, and the advancement of work was often constrained by cross-functional collaboration.
The team tried to combine AI with the open-source CAD tool to allow the system to read the 3D model, understand the product structure and assist in the generation of patented maps such as blast maps, parts labelling, etc. AI is responsible for understanding people ' s intentions, identifying the structures in the model, and then calling on engineering software for specific processing. It does not create a picture that looks "right" in the air, but works around real models.
This distinction is important. The patented drawings must respond to the actual structure of the product and not seek only visual integrity. AI can reduce pre-processing and communication, but the results are still subject to professional scrutiny, and the final quality cannot be judged by models alone.
The team ' s practice assumes that part of the pre-working period, which would have required a longer communication cycle, would have had the opportunity to be reduced to dozens of seconds or minutes. However, this does not mean that the entire process from product models to formal patent applications can be compressed to a few minutes, and engineering validation and professional clearance remain indispensable.
After the initial results of the tool, the team showed the boss Zhao Zhao Zhou. Whether or not to do a product starts with the organization of an AI strategy meeting at the corporate level is precisely the characteristic of winning the ongoing AI transformation: innovation begins not with a top-down planning document, but with a specific, headache in business processes.
The original attempt was not translated into a formal project, according to Cong Zelong, but to identify needs within the department, to do it first and to discuss the whole process.
This has given the team some space for error. Instead of proving that an AI project is worthy of investment and then applying for resources, it would be better to find problems in real work, to make something operational and to see if it was worth it.
When the instrument became operational, Zhao saw its commercial potential. Such a tool, along the lines of traditional software, which would solve practical problems, seemed to be available at a fee. However, a different consideration is suggested by the fact that patents cover research and development information that the enterprise has not yet made public and that potential users may be concerned if the tool must be uploaded; and that open sources and support local deployments offer more opportunities to lower the threshold of use by the enterprise.
One of the founders, who had crawled over the manufacturing industry for many years, first saw the commercial possibilities of the product; a young team leader wanted to put the product into a larger ecology. The two judgements meet in a specific product decision and ultimately promote the selection of the open source.
As Zhao Cheng subsequently explained his judgement: “If the business model of the traditional SaaS software enterprise were followed, it would also work. But he also knows that once the winner is determined to push AI on the ground, there will be a proliferation of AI-based product innovation cases like the patented map tool, and that AI Native will be calling out, “The open source is our moat, and this team and this path will be more valuable in the future, and the fee will be self-defeating.”
Of course, the open source of the product does not automatically amount to a moat, and whether there will be more users, external feedback and continuous overlap in the future will depend on whether the product itself really solves the problem. But this choice means, at the very least, that the goal of the team is no longer just to complete an internal tool, but to try to translate the experience accumulated internally into a capability that can also be used externally.
From Making a Tool to Building Toolbox
Following the patented charting tool, the success is also promoting other AI applications, one of which comes from recruitment.
The team developed an “AI test” tool that I hope to help companies use to screen candidates who meet the definition of AI Native talent and plan to open up to outside sources in the future. The AI Lab Network is also being developed to focus on the tool portals, downloads and use guides developed by the ICA Innovation Team.
These attempts have one thing in common: they do not identify a market-oriented product before looking for a picture of its use, but rather they grow out of the specific demand within the competition. Patent-based mapping tools correspond to R & D, legal collaboration, AI test counterpart screening, and other management efficiency tools attempt to address duplication of effort in day-to-day work.
But good internal tools may not be directly external. The data structure, workflow and security requirements of the different enterprises are not identical, and the open source does not mean that they can be used as soon as they are downloaded. More concerned is whether the tool, publish, will be able to obtain real feedback and continue to improve on the basis of its use.
This is also the reason why winning is prepared to bring together the relevant tools. If each tool is scattered among different teams, other staff members do not know where to look and outside users have difficulty understanding what products are available. The problem of distribution and use after the tools have been developed is one that the official network and the guide to use are trying to address.
From this perspective, success is not just a list of tools, but rather a path from business needs, from internal trials to external validation.
However, the number of tools does not directly demonstrate the effectiveness of the transition. The real question is whether the tool is going into day-to-day work, whether there is a willingness to continue using it, and how much time it saves and what improves. These questions still need to be answered in subsequent practice.
AI is changing management mode.
If the patent-based mapping tool shows how AI addresses specific business issues, then the internal “Hands of Hope” and “Stands of Hope” are different: As the threshold for tools falls, there will be more demand on the part of enterprises, but how to manage them will also be a new issue.
The team will collect the AI needs from the departments and assess them before deciding how to proceed. According to the information received, about two or three needs per week in the “gift pool” are to be advanced, “the need is not always met immediately, the team is limited in staff and there is a need to determine which issues deserve priority input”.
The success has benefited from this mechanism, which has transformed AI applications from exploration by a small number of technical staff into work that can be initiated, fed and engaged by other sectors. Staff do not need to have sophisticated skills before they have the opportunity to try to improve the process; they can ask questions first and then work together to find solutions.
But when there was so much need, new questions arose: what were the needs worth doing? Are the tools made useful? Can similar needs in different sectors be reused? If each issue is redeveloped once, AI may instead incur additional maintenance costs.
This contradiction coincided with the explorations carried out around Skill.
Skill can be understood as a set of instructions, processes and knowledge that allow AI to duplicate certain types of tasks. Since the end of last year, the Spectrum AI product teams have focused on this concept, and since March this year, they have been producing Skill for different departments and colleagues, “in the hope that some working methods will sink down so that staff will not have to start from scratch”. In practice, a number of problems were soon discovered.
Like the time taken to produce a Skill, sometimes exceeding the time saved by its use; the tools also need to be maintained on a continuous basis, and the frequency and actual effects are not easy to track; if each employee produces a single set, it is easy to repeat and it is difficult to judge what is worth retaining.
As a result, the team began to produce the SkillHub, hoping to bring the decentralized Skill together for easy search, reuse and management.
It sounds like a natural solution, but it's not easy to actually land. When should Skill be called? Can users judge its quality? If not, should it be optimized by developers or users? How can documents, the knowledge base and Skill be organized so that the accumulated experience can really be called upon when needed?
There are no available answers to these questions. It also reveals an easy-to-neglected fact in corporate AI: AI lowers the threshold for producing tools and organizing knowledge without automatically eliminating the costs of organizational collaboration. A tool can be made quickly, not to say that it can be maintained for long periods; or that a knowledge is preserved or that it will be used by the right person at the right time.
For an enterprise, AI is more than just an additional software, which means reconsidering the workflow process, the sedimentation of knowledge and the allocation of responsibilities. The easier the tools are to be developed, the more the enterprise needs to develop mechanisms to gauge demand, manage tools and measure impact.
Unlike Cong Zelong, who moved from specific tools, Zhao is more concerned about the ability of AI to change the core operational efficiency of its success.
Although Zhao is not a high-frequency user of AI tools, he tells us that he spends a great deal of time every day through short video media to study the latest AI research findings in the country and abroad, and then go to the premises of the Cong Zelong team to discuss, share and so on, and study what results can be applied to the practical work of successful companies.
"The focus of my concern is the AI effect of the R&D." In Zhao’s view, successful product R & D and innovation are highly dependent on the experience of engineers, data validation and complexity. If AI succeeds, it will shorten the R & D cycle and scale up success rate, which is the real productivity increase.
As a hardware enterprise, it develops a product that not only completes a design or a code, but also links electronic programs, metaware, purchase prices, supply risks, production assembly and certification. A change in one chain could affect the cost and progress of the entire product.
In the case of a charger, for example, from the product demand file (PRD), electronic engineers need to select core components and evaluate parameters, and procurement staff need to judge prices, and supply chains need to verify delivery cycles and cut-off risks. Once the programme has entered the assembly and validation phase, the team also assesses whether the cost is in line with the target. If the cost exceeds expectations, the pricing may need to be adjusted or the programme redesigned.
This is a chain that relies on experience, data and cross-sectoral collaboration. Even if it is not complicated to look at each link alone, it may take a lot of time to tie them together.
As described by the winning team, it is not uncommon for programmes to communicate, assess and adjust over time, with a one-month or longer cycle.
The problem is that many experiences are not fully embedded in the system. Engineers know whether the parameters of a metaware are appropriate and may remember what problems have arisen in the past by a supplier, but these judgements are often dispersed between individual experience and different data. Changes in personnel, switching of products or replacement of tools may make it difficult to replicate old experience.
Winning is not starting from scratch. It is understood that since 2015, it has started using product life-cycle management systems (PLMs) to build up material information, design specifications and risk records. On the basis of these data, the team built a database of metaware devices and connected to third-party and original plant platforms, synchronized metaware interfaces, voltage protocols, specifications, etc.
With these foundations, AI has the opportunity to participate in more complex R & D tasks.
At present, the Zelong team is exploring the full use of AI to make an Agent remote handle. It is envisaged that the user will take the handle and circle the content on the computer screen and interact with the voice. In the context of this product, the team attempted to establish a continuous AI R & D process for the definition of the need, the type of the device, the design of circuits and structures, the development of solids and upper machines, as well as the piloting and complete validation of the prototype.
The team has completed the selection of the AI device and the initial stacking of the structure, and AI has also generated the rationale and preliminary PCB Layout. At the moment, the team is working with Agent to review and evaluate the engineer, to control the data manual of the device, to check the selected interface, the layout around the antenna and the wiring constraints, and to adapt the design to the problems identified.
Next, the team will also need to advance the plate, welding and electrical debugging, which will be followed by a complete connection of external-driven and hardware, communications to the upper responder and so on, so that the design in the screen will gradually become a practical prototype.
But this is still a practice that needs to be validated. The completeness of meta-ware data, the timely updating of vendor information and the reliability of model judgements affect results. Hardware products are subject to physical testing and professional review, and AI offers programmes that do not circumvent these links.
Zhao Guocheng called this direction from "Vibe Coding" to "Vibe Engineering." He looks forward to extending the "demand, allow AI to write code" approach to hardware products to allow AI to generate programs, to operate engineering tools, to check and modify design, and to be judged and boarded at key nodes. The experience accumulated by manufacturing enterprises thus serves as the basis for guiding AI work, reducing the distance from one idea to one hardware product.
The division of labour between old guns and younger teams
Promising AI exploration ultimately falls on two different working paradigms.
Zhao has many years of experience in hardware manufacturing and is familiar with products, supply chains and business operations. He followed the latest developments in AI and asked whether those technologies could really solve the business problem.
The Zelong team is closer to implementation: needs are identified from the process of staff complaining, tools are produced and problems in use are brought back for adjustment.
The two are not simple “boss pay, young people do things”. Whether or not the patent-based mapping tool is charged and open-source is a discussion around product routes. Zhao has seen the possibility of commercialization, and Jun Zelong has put forward security, deployment and ecological considerations, and ultimately the company has chosen to open up its tools first.
Zhao considers the team's attempts on AI to be a long-term input, and it seems to him that the success of the team is a solid push for AI to land, and that AI Native is likely to be formed as a result of the proliferation of products of relevant innovative tools.
It is not just employees who use prompt or own a few tools. More importantly, can enterprises reorganize their work: The chain, which would have relied on individual experience, has been reduced to a reusable process that allows staff to identify problems more quickly, to test them and to verify them through real use.
There are a number of conditions that support such an attempt. It has real product R & D processes, accumulated material and design data for many years, as well as supply chains and production chains as applications. Rather than discussing in abstracto what AI can do, the team can put the tool into practice and see if it works.
Other enterprises may not have done the same, because products such as the patented map-based tool have been born to be successful, with a sense of admiration and a sense of priority and direction for AI creators, and from a first-place perspective.
It has been learned that, after Zhao's success, either the overall AI transformation of the winning company or the sharing, evaluation, recruitment and appraisal of the AI, is dominated by the Jun Zelong sector. It was also stated that in the course of an AI seminar, an innovation event organized within the Psychics, the company would award an AI application to all shared colleagues. At the present time, success is not only willing to try more in the direction of AI innovation, but also to open up some of the results to more people for testing.
This road is not without risks. In their exchanges, Cong Zelong also considered that some internal tools were only available for their own processes; that the needs raised by some sectoral colleagues did not necessarily warrant a productization; and that even open-source projects that lacked user and maintenance capacity could also be stalled. Even if the AI test, the SkyllHub and the patented mapping tool have produced initial results, it cannot be concluded that the firm has completed the AI transformation.
Promising is still facing more specific problems: Whether the AI tool is being used on an ongoing basis, whether external firms are being deployed successfully, whether there are measurable improvements in R & D processes, and whether the time saved can be translated into shorter product cycles and higher R Z success rate.
For Zhao Zhao Zheng Cheng, AI is ultimately going to fall into productivity; and the team will test how one-time, concrete innovation can be turned into a reusable and continuous renewal.
Right now, the winning Ai deformation is far from coming to conclusions. It may not be the amount of tools that have been put in place, but rather the fact that a traditional hardware company is beginning to try to change the way it solves the problem: so that innovation does not always wait for the formal setting, so that young teams can start from the field, and so that what is being made internally can be used and tested more widely.
Starting with a patented map, it is an attempt to build a path linking real needs, tools development, organization collaboration and external validation. Whether or not it becomes new competitiveness will ultimately have to be answered by the effects of its use. (This article was first sent to the titanium media app, in cash)
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.