AI 工具越来越多，企业推进 AI 的方式也在不断变化。但从企业管理者和一线实践者的视角看，AI 落地并能简单地采购工具、开通账号，再等待效率自然提升。
在 10 月 22 日举行的 InfoQ AI 侦察团 @QCon 上海 2026 活动中，四位侦察团成员分享了各自企业的 AI 实践进展、面临的挑战与下一步计划。有的企业已将 AI 融入日常工作，另一些则开始思考如何推动 AI 从个人提效走向组织协同，还有的企业正在探索客户沟通数据整合、财务报销自动化等具体场景。
这些真实具体的实践，给我们提供了观察企业 AI 落地的多维视角。
1 从个人提效到组织协同，AI 落地如何更进一步？
在研发实践中，AI 正在改变个人完成工作的方式。
来自 TGO 鲲鹏会的朱凯迪在采访中提到，一个人可以借助 AI 串联多个系统、并行推进任务，减少过去跨系统、跨团队协作中的部分沟通摩擦。他还分享了利用 AI 重构早年搁置的 C++ 游戏引擎项目的尝试：在提前梳理项目结构、技术要求和性能约束后，他使用 AI 在几个小时内推进了原本因工作量较大而搁置的项目，完成效果也让他感到满意。
但个人实践的效率提升，并不意味着组织可以忽略基础建设。他特别关注 Token 消耗与生产力之间的平衡，以及企业知识库的建设。如果团队缺少统一、可靠的知识基础，重复踩坑的问题可能从个人层面扩散到整个组织。
这也引出了另一个问题：当少数人已经掌握高效的 AI 工作方式，企业如何让更多人参与进来，并把有效经验沉淀为组织能力？
闯奇信息科技董事长黄炳川龙在采访中谈到，自己过去更关注个人或少数成员如何用好 AI，而下午的分享让他进一步思考如何推动组织整体提效。他希望下一步不再局限于少数人的使用，而是让更多成员参与 AI 组织建设，并通过研发效率等指标观察实际效果。
两人的关注点有所不同，却都指向一个转变：企业需要从“谁在使用 AI”，进一步走向“组织如何围绕 AI 重新协作”。
2 有知识库，不代表知识已经成为生产力
对很多企业而言，知识并不一定缺失。真正的问题是：它们是否被整理、连接，并能在需要时被有效调用。
正耀电子科技联合创始人、副总葛占国提到，企业已在财务、技术支持和软件移植等部门开展 AI 实践。在他看来，推进 AI 落地时，一个重要阻力是固有思维和工作方式难以改变。技术方案可以逐步搭建，但要让 AI 真正融入工作，还需要推动人的认知与习惯发生变化。
千零软件创始人、TGO 鲲鹏会上海会员陈冲，则已经将 AI 深入许多主要的工作流程。他认为创始人对 AI 这块的认知和投入是非常关键的，多数企业管理者未充分意识到 AI 转型的必要性，对投入产出存在顾虑，不敢放开投入尝试。同时他也提到一个具体问题：客户沟通数据分散在不同平台，限制了 AI 对完整信息的调用。他也在考虑如何把 AI 用于财务报销等职能流程，减少人工整理和核对的工作。
这两种实践都说明，知识库并不是一个孤立的技术项目，只有当信息能够被组织、连接，并进入具体的工作流程，知识才有机会真正转化为生产力。
3 效率之外，还要看投入、风险与组织机制
企业推进 AI 时，效率往往最容易被看见，但它并不是唯一需要衡量的指标。
朱凯迪提到，模型使用成本与 Token 消耗需要和生产力收益一起考量；陈冲则提醒，企业既要关注创始人对 AI 的认知和投入，也要重视企业知识体系的延续。如果关键经验过度集中在少数员工身上，即使 AI 提升了效率，企业也可能面临新的知识与人员风险。
从这些实践中可以看到，AI 落地至少涉及几个相互关联的层面：
- 投入是否值得： 不仅关注工具费用，也要考虑员工使用方式、知识准备与流程调整所需的成本。
- 流程是否适配： 将 AI 放入既有流程之前，需要先识别重复劳动、信息断点和真正值得自动化的环节。
- 知识能否延续： 让关键经验不只存在于个人脑中，也不只散落在不同系统里。
- 组织能否跟上： 技术应用需要与员工习惯、协作方式和管理机制共同调整。
这些问题很难靠单一工具解决，也很难通过一次培训完成。企业需要在具体场景中持续试验，根据效果调整应用方式。
4 下一步，不必一次解决所有问题
四位成员所处的 AI 应用阶段并不相同，但他们的实践都从具体问题出发：重构一个项目、推动更多员工参与、提升客户响应效率，或打通分散的业务数据。
这些具体的行动更容易形成清晰的验证路径：先明确一个值得解决的问题，梳理现有流程与数据基础，再决定 AI 应该承担什么任务、哪些环节仍需要人工判断，以及用什么标准衡量结果。
企业 AI 转型的难点，是持续回答一系列具体问题：哪些工作值得改变？改变之后是否更有效？经验能否复制？风险是否可控？AI 没有替企业做出所有判断，它改变的是企业解决问题的工具与方式，而选择解决什么问题、如何组织资源、怎样验证效果，依然需要企业自己回答。
InfoQ AI 侦察团希望创造的价值，是让正在探索 AI 落地的企业决策者有机会走到一起，通过真实的案例、坦诚的讨论和跨企业的经验碰撞，共同分析并拆解这些具体问题，帮助企业少一些盲目试错，多一些有依据的判断，让企业决策者带着更清晰的方向回到业务现场，继续验证、迭代，推动 AI 从试点走向真正的业务实践。
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。
The number of AI tools is increasing and the way businesses advance AI is changing. However, from the perspective of business managers and front-line practitioners, AI landed and could simply purchase tools, open accounts and wait for a natural increase in efficiency.
During the InfoQ AI Reconnaissance Mission @QCon Shanghai 2026 held on 22 October, four members of the Reconnaissance Mission shared information on the progress of their businesses in AI practice, challenges faced and the next step plan. Some companies have integrated AI into their day-to-day work, others have started to think about how to move AI from personal effectiveness to organizational synergy, and others are exploring specific scenarios such as customer data integration and automation of financial claims.
These are real, concrete practices that provide us with a multidimensional perspective to observe corporate AI landings.
1 How can AI go further from personal effectiveness to organizational synergy?
In research and development practice, AI is changing the way individuals perform their work.
In an interview with TGO Zhucadi, it was mentioned that an individual could use AI to link multiple systems and advance tasks in parallel to reduce part of past communication friction across systems and teams. He also shared an attempt to re-engineer the C++ engine project using AI to re-engineer the early years: He was also satisfied with the results of the completion of the project, which had been put on hold because of its heavy workload, using AI in a few hours.
However, the increased efficiency of individual practice does not mean that the organization can ignore infrastructure. He paid particular attention to the balance between consumption and productivity in Token and the building of an enterprise knowledge base. In the absence of a unified and reliable knowledge base for teams, the problem of repeat pitfalls can spread from the individual level to the entire organization.
This also raises another question: how can a company involve more people when a few are already mastered of efficient AI working methods and sink effective experience into organizational capacity?
In an interview with Huang Bingchuan, chairman of the board of directors of Rudd Information Technology, he said that he had in the past focused more on how to use AI by individuals or a few members, and that this afternoon ' s sharing had allowed him to think further about how to make the organization work as a whole. He hoped that next step would no longer be limited to the use of minorities, but would involve more members in AI organizational development and would see the practical effects through indicators such as research and development efficiency.
Both have different concerns, but both point to a shift: business needs to move from “who is using AI” to “how to re-engage around AI”.
2 Knowledge base does not represent knowledge becoming productive
For many enterprises, knowledge is not necessarily missing. The real question is whether they are organized, connected and can be effectively called when needed.
The co-founder and vice-president of Enlightenment Electronics Technology mentioned that enterprises have been implementing AI practices in such sectors as finance, technical support and software transplantation. In his view, an important resistance to advancing AI's landfall is the innate difficulty of changing thinking and working methods. Technological programmes could be built gradually, but for AI to be truly integrated into the workforce, it was also necessary to promote changes in human perception and habits.
The founder of the thousands-zero software, the TGO Zhui Sea Member Chen Zheng, has advanced AI into many of the major workflow processes. He considered that the founders ' awareness of and input to AI was crucial, and most business managers were not fully aware of the need for AI to transform, were concerned about input output and were afraid to let go. He also referred to a specific issue: client communication data were scattered across different platforms, limiting AI ' s access to complete information. He was also considering how to use AI for functional processes, such as financial claims, to reduce manual processing and reconciliation.
Both practices demonstrate that the knowledge base is not an isolated technology project and that knowledge has the opportunity to be truly productive only if information can be organized, connected and entered into a specific workflow process.
3 In addition to efficiency, input, risk and organizational mechanisms
While efficiency is often the most visible when enterprises advance AI, it is not the only indicator to be measured.
Jucedi mentioned that the cost of using models is considered in conjunction with Token consumption needs and productivity gains; Chen is a reminder that enterprises need to focus both on the awareness and input of their founders to AI and on the continuity of their knowledge systems. Even if AI improves efficiency, enterprises may be exposed to new knowledge and human risk if key experience is over-focused on a small number of staff.
As can be seen from these practices, AI landed on at least several interrelated levels:
- Whether the input is worth paying attention not only to the cost of the tool but also to the cost of staff usage, knowledge preparation and process adjustments.
- The adequacy of the process: Before putting AI into an existing process, it is necessary to identify duplicates, information break points and links that are truly worthy of automation.
- The continuation of knowledge: let the key lessons not only exist in the individual brain, but also in different systems.
- The organization ' s ability to keep up: technology applications need to be adapted with staff habits, collaborative approaches and management mechanisms.
These problems are difficult to solve by means of a single tool or through a single training exercise. Enterprises need to continue testing in specific scenarios and adapt their application to their effects.
4 next step, no need to solve all problems at once.
The AI application phase was different for the four members, but their practice proceeded from a specific point of view: reconstructing a project, promoting more staff participation, improving client responsiveness or creating decentralized business data.
These specific actions make it easier to develop clear verification pathways: First, identify a problem that deserves to be addressed, combe the current process and data base, then decide what AI should undertake, what links still need to be judged manually, and what standards to measure results.
The difficulty of the transformation of AI is to continue to answer a number of specific questions: What is worth changing? Is change more effective? Can experience be replicated? Are risks manageable? AI does not make all the judgements for the enterprise; it changes the tools and the way in which the enterprise solves its problems, chooses what to solve, how to organize its resources and how to validate its effects, and it still needs answers from the enterprise itself.
The InfoQ AI Reconnaissance Mission seeks to create value by providing an opportunity for business decision makers who are exploring AI to come together to analyse and decipher these specific issues through real cases, frank discussions and cross-business experiences, to help businesses to test them less blindly, to make more informed judgements, to return business decision makers to the field in a clearer direction, to continue to validate and to evolve, and to move AI from pilot to real business practice.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.