Second-generation factory leaders take the helm, using Feishu AI to propel their factories to "Wuhu" success
According to the organizer's article, a group of Feishu AI users brought real workplace and daily-life questions to its Beijing event on July 25.
Authentic Insights from Frontline Practitioners
During the first half of the event, six frontline practitioners took the stage one by one to share first-hand accounts of their AI application cases.
Midway through the presentations, a young man walked onto the stage wearing a safety helmet and canvas shoes. He is Xiao Pan—the "second generation" leader of an auto parts factory in Wuhu, Anhui province—affectionately called the "young boss" by his coworkers.
According to Xiao Pan, after returning home to the family business, his first formal AI project used AI with cameras to help identify situations in which workers were not wearing safety helmets. His story begins with that very helmet.
From "Wuhu Takeoff" to "The Sky Is Falling"
Before returning to the factory, Xiao Pan had dreamed of effortlessly becoming a wealthy second-generation heir, watching his business "Wuhu" take off. But reality shocked him: "When I came back, I felt like the sky had collapsed."
What he saw was this: operations still relied on paper-based documents; repetitive tasks in Excel were handled manually; employees’ understanding of AI mostly stopped at seeing it as just a chat tool.
So he positioned himself as a Field Deployment Engineer (FDE): understanding factory workflows while turning on-site problems into tasks executable by AI.
Bringing AI into Factories: No Room for Armchair Theory
Xiao Pan quickly realized that introducing AI tools was only the first step. The real challenge lay in integrating AI into employees’ daily routines.
How could he encourage AI-novice employees to give it a try? How could he help them understand the boundaries of AI capabilities? And how could individual explorations evolve into plant-wide habits?
From data digitization and embedding AI agents into chat groups, to establishing training and incentive mechanisms, Xiao Pan gradually carved out a practical path for AI adoption in traditional factories.
Digitalize First, Then Intelligentize
In the actual transformation process, Xiao Pan didn’t rush to develop intelligent agents. Instead, he prioritized solving foundational data issues.
By the end of 2024, he used AI-assisted coding to build small programs that automatically completed repetitive Excel tasks, and gradually migrated paper-based process cards and other forms into Feishu's multi-dimensional tables.
In his view, the prerequisite for a factory to embrace AI is achieving digitization and information integration. Multi-dimensional tables transformed scattered information from paper and spreadsheets into dynamic data streams, enabling subsequent AI agents to read content and track processes.
Therefore, the first step in AI implementation isn’t generating answers—it’s getting real operations online.
Filling Talent Gaps Matters More Than Improving Efficiency
Xiao Pan believes that most small and medium-sized factories today face a more urgent need than efficiency improvement: filling long-standing vacancies in specialized roles.
His chosen starting point was quality management. According to his account, the factory needed to align relevant automotive-parts quality work with IATF 16949 requirements but had struggled to hire a suitable quality-systems engineer, so he decided to build a "Quality Expert" AI agent.
He designed the agent around three core functions:
Review documents: drafts checks and flags potential issues against industry standards for human confirmation;
Answer questions: employees receive draft responses without first searching through manuals, while professionals review critical matters;
Train newcomers: handle basic inquiries so senior staff can focus more attention on complex tasks.
According to his account, many shop floor workers now consult the AI agent first when problems arise. It offers preliminary reviews and suggestions, while critical decisions regarding compliance and quality are still reviewed and confirmed by professionals.
He gave one example in which R&D staff submitted system documentation for feedback in a group chat. In his account, the "Quality Expert" generated a review report, listed proposed corrections and drafted a revised version for colleagues to verify.
In this collaborative model, the AI agent performs initial screening, professionals make final judgments, and the process is designed to be traceable and double-checked, feeding into the quality management workflow.
Embedding AI Agents into Messaging Groups
After building the agent, Xiao Pan didn’t hide it behind a separate interface. Instead, he embedded it directly into the instant messaging groups employees use every day.
When employees ask questions, the AI responds based on its knowledge base, while domain experts can chime in to supplement or correct answers within the same thread. According to Xiao Pan, discussion histories are preserved and the AI periodically summarizes frequently asked questions and high-quality cases, which are then fed back into the knowledge base after human review.
This approach speaks to a longstanding knowledge-transfer problem in traditional factories: much craftsmanship and experience resides in the minds of veteran workers and can be lost once a problem is solved. With human review, records of human-AI collaboration may instead be retained as reusable organizational knowledge.
Old expertise becomes contextual input for the agent’s future work.
Meanwhile, because the agent is publicly visible in group chats, its capabilities and limitations are transparent. When answers are accurate, trust builds; when errors occur, experts can correct them promptly. This open interaction also allows R&D, quality, and management teams to stay aligned on issue progress and outcomes in one shared space, reducing cross-departmental information gaps.
Lowering Barriers to Drive Full Participation
Despite having advanced from paper forms to AI agents in messaging groups, Xiao Pan still faced the most fundamental question: Even if the tools are ready, will employees actually use them?
As the driving force, he needed not only to solve problems and build tools, but also to lower the usage barrier enough for frontline workers to be willing to try.
In small factories, unused tools are common. Xiao Pan summarized three key insights:
First, tools provided by leadership must be plug-and-play—ideally activated with a single command. Any noticeable barrier may deter users. In his words, they must be “foolproof.”
Second, tools developed by peers spread faster than those mandated from above. Employees are more likely to think: “If my coworker can build it, maybe I can too.”
Third, mechanisms should be designed top-down, but behaviors must emerge bottom-up. Management provides resources, training, permissions, and incentives, while employees identify use cases, create tools, and share results.
To support this, he plans to build an enterprise skill library, turning validated AI tools into shared resources for all, and using monthly usage leaderboards and incentive programs to spotlight effective tools.
He also admits that while leaderboards reflect usage frequency, they don’t fully capture actual value creation.
Tools Are Just the Beginning—Organizational Readiness Is Key
At this stage, Xiao Pan gained deeper insight into AI implementation: success shouldn’t be measured merely by how many AI agents were built.
More important is the organization’s supporting capacity: whether training is adequate, data is connected, responsibilities are clear, and individual experiences can be turned into reusable team methodologies.
Tool delivery is just the beginning. "I’m merely lighting a spark. Most of the actual work still needs to be carried out by employees themselves."
Getting factory workers comfortable with AI cannot rely on a single training session or demo.
Xiao Pan’s approach is clear and pragmatic: first, use Feishu multi-dimensional tables to turn paper-based information into dynamic data; then introduce AI agents into chat groups for employee queries and expert validation; have AI summarize full interactions, followed by human review to enrich the knowledge base; finally, drive broader adoption and tool creation through training, a skill library, and incentive mechanisms.
Xiao Pan lights the spark; in his telling, Feishu links data, communication, AI agents and knowledge accumulation into one working workflow.
This transformation, sparked by a single safety helmet, aims not at creating a smarter boss—but at building an entire factory capable of implementing AI with Feishu.