Companies use Feishu to reorganize information and work, and the context that accumulates from this helps improve company efficiency.
Article by Shen Xing
A large model can perform completely different tasks in different companies.
It can write reports and summarize meetings, but if it doesn't understand what the company is doing and what decisions have been made in the past, it is still difficult to truly engage in work. Many tasks in companies are not completed with a single command. For AI to continue its work, it needs to know where things begin, what has happened in between, and why things have become the way they are today.
The automotive industry can especially illustrate this issue.
Behind a car is a complex project completed by demands, software, hardware, testing, manufacturing, and supply chains. A change in a feature can simultaneously affect the software at the vehicle end, mobile phones, cloud services, chips, and production lines. The more people and companies involved, the faster the research and development pace, and the more important it is to connect the information.
In the past, this information was primarily for reading, communication, and judgment. After large models began participating in research and development and operations, they gained a new layer of function.
In the development process of automotive companies, a feature goes from the demand stage to final delivery, often undergoing multiple reviews, development, testing, and modifications. For example, when the assisted driving function expands from highway scenarios to urban roads, AI not only needs to understand the latest demands but also needs to know previous user feedback, testing data, hardware changes, and unresolved safety issues.
When a digital key cannot be used on some mobile phones, AI cannot just look at a fault report; it needs to simultaneously track records from the phone, vehicle end, cloud, and hardware to determine where the issue lies.
For AI, these demand changes, testing results, fault records, and supplier communications are the context for understanding a project. Only by linking the causes of problems, the solutions that have already been attempted, and the unresolved limitations can it provide executable judgments for the next steps.
Moreover, today's new changes in the automotive industry make this matter even more urgent. The variety of models is increasing, software updates are becoming more frequent, and the teams, systems, and suppliers connected to a feature are also multiplying. The information transfer that used to rely on repeated human communication is becoming harder to keep up with the speed of research and development and delivery.
Since August, we have interviewed more than five automotive manufacturers and supply chain companies. Baolong Technology needs to connect OEMs and both German and Chinese research and development teams simultaneously; Yinke Technology's digital key needs to span mobile phones, vehicle ends, cloud, and hardware production, while also adapting to the different developmental systems of various automakers.
While the specific problems faced by these companies vary, the common issue is: information is scattered across different companies, teams, and systems. For AI to participate more deeply, the first step is to reconnect this dispersed and constantly changing information.
Some automotive companies recognized the importance of context very early on. In May 2021, Li Xiang, founder of Li Auto, referred to Feishu as a "super perception tool" at the Feishu Future Unlimited Conference. In his view, what flows through Feishu is not just office and information but also the cognition and knowledge within the company, connecting them through collaboration.
Beyond Li, the use of Feishu in the automotive sector is no longer limited to new force car companies. From vehicle manufacturers like SAIC to supply chain companies like Bosch, Linglong Tire, and XWANDA, it has entered the automotive industry chain alongside complete vehicles, components, tires, and power batteries. These companies have begun to汇集 scattered work information from documents, messages, meetings, spreadsheets, and processes into the same collaborative environment, allowing a task's development to be continuously preserved and giving AI the opportunity to understand work and enter processes.
Feishu CEO Xie Xin also stated at the AI Productivity Conference's automotive session that the vast majority of tasks in companies do not occur in isolated tools but span documents, messages, meetings, spreadsheets, processes, and collaboration between people. When agents enter the company, the work mode will shift from "people using software" to "people and agents completing tasks together in the same work environment."
In a factory, there is a production line for automobiles. Around AI and the internal and external context of the company, the automotive industry is forming a "second production line."
Feishu aims to enter precisely here.
How "Context" Changes Cooperation Models in the Automotive Supply Chain
In the complex automotive industry, context often does not exist solely within a single company.
The research and development of a vehicle involves OEMs and numerous suppliers. In the past, OEMs would propose requirements, suppliers would complete component development, and then deliver the finished product to the OEM for integration. As automotive electronics and software have become increasingly important, both sides have begun to jointly design and develop. OEMs now engage earlier in software and system definition, and suppliers are entering vehicle architecture earlier.
If one party modifies its requirements, the design, testing, and delivery pace of the other party may also change. The originally scattered development processes across different companies are becoming more closely connected, which makes the context needed in the automotive industry naturally cross corporate boundaries.
Business has become interconnected, yet information is still scattered across different systems. Both OEMs and suppliers have their own requirements and project systems, and communication still relies on Word, Excel, email, and meetings. Documents can be sent, but the relationships between requirements, designs, development, and testing are hard to carry over together. After resolving an issue in a meeting, both parties must return to their respective systems to update their statuses.
As a result, project managers spend a lot of time on information and pushing for progress, often requiring manual explanations for why the same issue displays different statuses in two systems.
For AI, all it sees are isolated information fragments. For instance, if camera parameters change, it may not understand that it will affect algorithm calibration, controller software, and vehicle testing; if a supplier is delayed, it may not be aware that it will hold up a specific round of test production.
In mid-2026, SAIC and United Automotive Electronics used Feishu to connect a research and development project. Originally, information conveyance, plan synchronization, change notifications, deliverables, and review materials scattered across emails and phone calls can flow directly between the two systems, with related documents, meetings, and discussions attached to specific matters.
SAIC incorporated design requirements, component design tasks, hardware and software development, testing, and review into 187 main process nodes in the same project chain. Each demand can continue to correlate with design, development, and testing tasks, clearly specifying responsibilities and timelines. Upon reaching a node, the system automatically reminds and summarizes, sparing project managers from needing to ask one-by-one, shortening the scheduling of all plans from one to two hours to five minutes, and significantly reducing the time needed for issue tracing and project handovers. "There are seven roles on this chain, belonging to two companies, all running on the same process today," said Wang Conghe, Director of the Innovation Research and Development Institute of SAIC Group.
More importantly, the reasons for changing requirements, who made decisions, and how one modification affected multiple links—previously largely stored in the memories of project managers and engineers—can now be preserved as context for AI to understand the business. “AI does not lack computing power; it lacks context," Wang Conghe said. When cross-enterprise tasks, schedules, deliverables, and decision-making processes are continuously preserved within the same process, AI can truly understand the research and development site. The pitfalls a project has encountered have the chance to become experiences that the next project can directly call upon.
However, the collaborating parties do not thereby enter the same internal system. SAIC and United Automotive Electronics belong to two different Feishu tenants, each retaining its own data, processes, and permissions. The OEM defines the jointly promoted tasks and delivery nodes, while suppliers maintain their internal processes, opening only the necessary data and statuses.
This aligns well with the reality of the automotive supply chain. OEMs and suppliers need to work together but cannot open all data to each other. Intellectual property, responsibility allocation, and data boundaries all require that information flow while maintaining boundaries.
The process by which Feishu enters the automotive industry is also changing Feishu in turn. It must understand APQP, IPD, ASPICE, and quality nodes in complete vehicle development while adapting to suppliers' data boundaries, the processes of different auto companies, and the division of labor between specialized engineering systems and collaborative tools.
In July of this year, SAIC and Feishu expanded their collaboration to the group level. Prior to this, SAIC GM, SAIC GM Wuling, Zhiji Automobile, United Automotive Electronics, and other companies were already using Feishu individually. After signing the group contract, both parties began extending cooperation to R&D projects, collaborative efforts between complete vehicles and parts, AI applications, and even more aspects like "research, production, supply, marketing, and service."
Similar collaborations are emerging in more automotive companies. Public data reveals that among the top ten brands of new energy passenger cars in sales for 2025, seven have chosen Feishu projects; approximately two-thirds of the top 30 new energy vehicle companies in domestic sales have already collaborated with Feishu. The market shares of Feishu projects in software R&D management SaaS and IPD management SaaS have reached 46.8% and 68.6%, respectively.
Zhang Jianqiong, Director of the Intelligent Components Department of SAIC Passenger Vehicles, likened this relationship to the early collaboration between IBM and General Motors: software companies first accompany their first customer through a complex project, then find reusable parts from it to create products that can also be used by other companies. The first customer incurs more trial and error costs, and the software company genuinely learns the industry through this.
As we enter the era of large models, this accumulation adds another layer of value. Whether a company can effectively use AI depends not only on the model itself but also on how much genuine, continuous, and clearly permissioned business context it can gain. Feishu's previous connections of projects, documents, meetings, and processes in the automotive industry thus become part of the foundation for AI to enter real business.
Internal AI and "Context" Practices in Automotive Companies
Aside from the context between supply chains, an automotive company has another set of even more complex information internally.
The variety of models is increasing, and development cycles are shortening. AI is beginning to enter research and development, quality, and management processes. AI within the company is also much more complex than an individual opening a chat box. It not only needs to understand a document but also needs to know which project the document belongs to, who is authorized to view it, which tests are related, and whether its answers can be used for formal decisions.
Meetings, documents, group chats, project nodes, quality standards, and historical issues collectively record how a company works. Over the past few years, many automotive companies have retained this information within Feishu while also maintaining the relationships between personnel, projects, and permissions.
The transformation of 四维图新 (Four Dimensions Map) is quite typical.
The company initially considered building its own RAG knowledge base and organizing corporate materials to hand over to AI. However, they later realized that corporate information changes daily and personnel and projects have complex permission relationships. Rebuilding a knowledge base nearly equated to duplicating an organization and permission system.
Ultimately, 四维图新 directly uses the existing chat, meetings, documents, and permission relationships in Feishu, allowing AI to read information within the original permissions. Feishu has already embedded itself into daily business, automatically updating these contexts with work, and there's no need for the company to maintain an additional set.
For manufacturing enterprises like 浙江黎明 (Zhejiang Dawn), context also includes another type of information: experience accumulated by senior craftsmen.
The company has organized experience in cold extrusion, machining, and mold design into its intelligent agents. For new design tasks, AI first provides suggestions, and engineers then modify it; design reviews are also first checked by AI, marking key points, and experts later make the final judgment, reducing review time for individual drawings from about three hours to 10 to 20 minutes.
The process experience, design rules, and judgment methods that were previously scattered in the minds of senior craftsmen have turned into capabilities that the organization can repeatedly utilize.
However, letting AI know what has happened in a company is just the first step. As agents enter the office scene, the new question is whether AI can utilize this context to continue with subsequent work.
Dosan Bean Work, released on August 25, provides a new way to connect. It can break down tasks around objectives, call tools, and continuously advance complex work. In an enterprise context, Feishu retains employee identities, permissions, organizational relationships, and work records, while Dosan Bean Work functions more like a execution framework, allowing models to use existing company tools and processes based on these contexts.
This also allows enterprise AI to continue working based on the company's real status.
Yinke's CTO Zhang Liang had already been attempting a similar approach. The departments, jobs, and roles involved in Yinke's integrated product and project are myriad. Understanding the latest progress of projects and products quickly to support the next decisions used to require a lot of time and energy; now he can organize meetings, track objectives, and summarize progress using multiple digital avatars. Meanwhile, because the project and meetings have already recorded the work process, he no longer requests the team to write separate weekly reports, thus dramatically increasing the efficiency of information retrieval. “We have come to believe that it can understand my context well,” he said.
However, enterprise context does not naturally aggregate in the same system, nor is all information suitable to be integrated. Yinke still has to communicate with customers through DingTalk, WeChat for Work, and personal WeChat; this information cannot automatically return to Feishu. Even if it is technically possible to connect, whether the information is willing to be open also depends on the interests and trust between enterprises. Baolong's CIO Huang Junlin believes that if OEMs only use process transparency to further pressure prices and compress cycles, suppliers will not be willing to open more information. Meanwhile, due to GDPR compliance, employee privacy, and internal company policies, European teams have set stricter boundaries on data and meeting records, which may also restrict certain capabilities.
Therefore, AI needs to understand enterprises, but "understanding" never equals seeing everything. The relationships in which information generates and where it can flow are inherently part of the context. Feishu has already accumulated a significant amount of project, document, and collaboration records from automotive companies, and what ultimately determines the value of these accumulations is whether they can further convert into tasks that AI can reliably execute within clear permissions and boundaries.
As Large Models Enter Various Industries, People Become More Important
The automotive industry is not an exception.
As AI moves from answering questions to participating in work, more and more industries are beginning to recognize the importance of context. The same large model can play a varying role in different enterprises, depending on the enterprise's own knowledge, data, permissions, and work records.
The AI assistant Morgan Stanley created for wealth management advisors connects the company's own research and knowledge base. Currently, it can answer questions from approximately 100,000 internal documents, with over 98% of the advisor teams using it. Decades of accumulated research, processes, and customer service experience enable AI to address real concerns of a financial institution.
Walmart is doing similar things. Its Wally aggregates sales, inventory, and demand data to help purchasing teams determine where items are out of stock and why; the employee-facing MyAssistant can read store, team, and operational information. Retail operations change daily, and AI can only participate in actual operations after accessing this continuously updated information.
However, the context within enterprises does not solely exist in documents and databases.
Codes, systems, and SOPs record the rules explicitly written down by the enterprise, but much of what truly affects work has remained with individuals.
A quality engineer knows that a certain type of anomaly occurred several times in the past and how it was resolved; a procurement head understands what the same delivery commitment means when placed on different suppliers; frontline employees know the two steps in the ten-step process manual that are most likely to cause problems.
Many of these experiences have not been fully written into systems but are affecting enterprises' decision-making daily. For AI to participate in these judgments, it must gradually comprehend the information previously scattered among people and departments.
This also means that after AI enters a company, merely integrating existing materials is insufficient. Management must decide which businesses are worth investing in, which data can be opened; business leaders need to clarify the real issues; and frontline employees are most aware of which tasks recur and which judgments rely heavily on experience.
Human participation itself continuously supplements the enterprise's context.
This is why, in the past year, many vendors have begun to focus more energy on the people within enterprises. Companies like Alibaba Cloud, DingTalk, and Microsoft have launched AI training, certification, and practical projects aimed at enterprise managers and business personnel.
Starting from 2023, Feishu has held AI competitions and other activities targeted at enterprise management and frontline employees intensively.
Zhejiang Dawn has hosted three AI Pioneer Competitions to date. The initial focus was on the novelty of ideas, but later it began to look at how much time was saved and what real effects were produced; in preparation for the fourth competition recently, they added "how many people are using it" as a scoring point. Meanwhile, the company has divided the competition into two tracks, one for office and one for production sites, providing a stage for more frontline employees to showcase their talents.
On the surface, these competitions are aimed at finding good AI applications, but they also gradually reveal a company's latent experiences and discover AI talents of the new era; some AI pioneers have showcased their results through the competitions and later moved into management positions.
Which tasks are the most time-consuming, which processes are long-standing and unreasonable, which positions hold knowledge that others are unaware of—these have often been most clear only to those who are part of it. When employees begin to use AI to address these issues, they also transform the experiences originally stored in their minds into methods that can be documented, replicated, and reused.
New technologies entering businesses often go through a process of a few people trying them out to changing the working methods of the entire organization.
In the late 19th century, when electricity first entered factories, many businesses merely replaced steam engines with electric motors, with no significant changes to machine placement and production processes. Later, factories began redesigning production lines around electricity, allowing machines to break free from central drive shafts and being rearranged according to production needs, thus enhancing efficiency.
AI may be undergoing a similar phase. A few early adopters using it only proves that it is useful; real change will come when a company starts reorganizing its knowledge, experience, and work practices.
In the past, the most important capabilities of a company were well-hidden among its people and organization. In the future, whether these experiences can be understood by machines and used by more people will gradually become part of enterprise capabilities.
Image source: "Ferrari."
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
