One person, one laptop, a Feishu chat window, and a Feishu CodeM development agent.
This article is reposted from the WeChat official account: Feishu Projects
On an ordinary workday in August 2026, a young employee at CATL (Contemporary Amperex Technology Co. Limited) picked up his tray in the cafeteria. Before sitting down, he took out his phone and opened the internal dating and social app “Qiaoqiao Hui.”
The homepage recommendation list used a warmer color scheme. AI suggested 30 matching results, with a new "View Details" button added beneath each card. The employee swiped through several pages, saved profiles of interest to his favorites—completed in under a minute.
The developer responsible for this update was a member of CATL’s IT team who simultaneously maintained multiple internal systems; “Qiaoqiao Hui” was just one of them. Collaborating with him on this iteration was an AI development agent called Feishu CodeM.
One Person Maintaining Multiple Projects Simultaneously
“Qiaoqiao Hui” was first launched in 2020 under the passive leadership of CATL's labor union. Its purpose was simple—to provide eligible internal employees with a trustworthy platform for matchmaking.
Consumer batteries are CATL’s core business. In recent years, rapid industry growth has led to workforce expansion, particularly among younger demographics. With numerous factories, tight shift schedules, fast-paced work environments, and the relatively small city size of Ningde, Fujian Province, young employees naturally have limited access to social circles. The original idea behind the app was to help colleagues get to know each other within the same workplace.
In its early days, “Qiaoqiao Hui” had basic functionality: registration, profile filling, browsing, and sending connection requests. Though usage numbers were low, there were always users. Some employees even found their partners through the app and reported back to the union—receiving only brief thanks in return.
By 2026, however, both account volume and daily active usage of “Qiaoqiao Hui” had increased significantly compared to previous years. Ahead of the Qixi Festival, the union hoped to make match recommendations more accurate and the interface design more suitable for romantic connections. Young employees’ aesthetic preferences and habits were also evolving, requiring updates to details such as color tones, field layouts, and date formats.
Logo of CATL’s Labor Union “Qiaoqiao Hui”
The requirements were passed by the IT team to the assigned developer. He wasn’t a dedicated developer for “Qiaoqiao Hui”—he also managed various internal systems, employee services, and multiple lightweight applications. His desktop always had several IDE windows open.
Managing multiple projects alone, each with clear launch deadlines—this is a real scenario commonly seen in many large enterprise IT teams. Business demands keep coming, but development resources remain extremely limited. Under traditional development methods, relying solely on manual coding, documentation review, and style adjustments, trade-offs must be made within limited time. In previous years, “Qiaoqiao Hui” would typically receive only critical bug fixes around this time, while smaller, “nice-to-have” optimizations were often postponed until the next quarter.
This time, however, the situation was slightly different.
Feishu CodeM Takes Over the “Minor Updates”
In early August, the developer installed Feishu CodeM on his laptop and opened a Feishu chat window. Instead of drafting a detailed requirement document, he directly typed one line into the chat: “Change the background color of the recommendation cards to a warm tone.”
Feishu CodeM accessed project code within its permission scope, located the corresponding front-end style file, and after making modifications, directly attached a screenshot of the updated page in the chat reply—showing both the code changes and final visual effect. The developer could instantly judge whether the color and layout met expectations, then decide whether to accept the changes.
He then issued the next set of instructions:
* “Standardize the date format to month-day.”
* “Reorder several fields on the personal profile page.”
* “Limit the AI recommendation list to a maximum of 30 items, sorted by MatchScore from high to low.”
* “Restore the ‘View Details’ button below both the AI recommendations and latest lists, while keeping the favorite button.”
Each request clearly specified a particular page, field, or rule. After Feishu CodeM implemented the changes, the developer reviewed them, ran local tests, and decided whether to keep or revise them.
At one point, he encountered an uncertain issue: How was the sorting logic originally designed for the AI recommendation list? Why was MatchScore chosen instead of other fields? The discussion was scattered across chat logs in the “Qiaoqiao Hui” requirement group, spanning nearly a year. In the past, when facing such situations, he would either scroll back through the chat history himself or directly ask colleagues involved in the earlier discussions.
This time, he granted Feishu CodeM permission to read messages from the designated group. Within the authorized scope, Feishu CodeM retrieved relevant discussions, identified key information fragments scattered across nearly a year of chat records, and provided contextual judgment and suggestions—all within the conversation. The developer could directly see cited historical messages alongside Feishu CodeM’s analysis, enabling him to determine which parts to adopt.
For the developer, Feishu CodeM handles more than just code. Project code, skill plugins, and reusable knowledge can all be invoked within the same workspace. For developers maintaining multiple projects, having code, context, and related tools centrally managed allows AI to truly understand the current task at hand.
Steps such as testing, acceptance, code merging, and deployment were not handed over to the tool. Every page, every navigation, and every recommendation result in the released version of “Qiaoqiao Hui” was personally clicked, reviewed, and confirmed by him. On the day before launch, he was still monitoring page loading speed and went back to adjust the caching strategy upon noticing a slow-loading avatar.
One person, one Windows laptop, one Feishu chat window, plus one Feishu CodeM development agent. Pre-release went live on August 12, with full promotion starting on August 19—the Qixi Festival. This timeline was set entirely by the IT team at their own pace.
What Was Truly Saved
Were the tasks previously deemed “too busy to handle”
When asked how much time was actually saved, the developer gave a modest answer. “The scope of these changes wasn't huge—perhaps about half the time was saved. That should have been the effect.”
But for specific modifications, the efficiency gain was very tangible. “You simply describe a certain page, a certain field, what content it should display and in what date format, and Feishu CodeM precisely makes the change for you.”
In the past, modifying a field’s display format required first locating the relevant component, identifying where the date formatting function was called, editing the code, running tests, and checking the outcome—a process that often took half an hour. With Feishu CodeM, from typing ‘standardize date format to month-day’ to seeing the correct display on screen, it now takes only a few minutes.
Even more significant changes occurred in small optimizations that were frequently postponed. These tasks weren’t difficult—but they shared a common problem: important to do, yet rarely prioritized due to time constraints.
After introducing Feishu CodeM, the completion rate of these minor improvements increased significantly. The developer remained the same individual managing the same number of projects, but now each project had greater room for refinement.
For an app serving internal employees, these details are far from trivial. Slightly smoother page interactions, clearer information presentation, and more accurate recommendations—all ultimately reflect in employees' actual user experience.
Wherever Requirements Extend, AI Steps In
The “Qiaoqiao Hui” case represents a starting point for CATL’s exploration of Feishu CodeM.
Throughout this process, Feishu CodeM did more than just modify code. It delved into a real enterprise development environment, understood project context, undertook specific development tasks, and collaborated with developers throughout the entire journey from requirement to code delivery. Developers still take responsibility for requirement evaluation, solution design, and code review, while AI handles repetitive development tasks suitable for automation.
This further revealed to CATL that the value of a coding agent goes beyond improving the efficiency of individual coding tasks. When AI integrates into existing enterprise development workflows and participates progressively alongside requirement evolution, developers can redirect more energy toward work requiring professional judgment.
In fact, CATL plans to use Feishu CodeM throughout the construction of its pre-manufacturing inventory management system. Compared to “Qiaoqiao Hui,” this system is closer to core business operations, involving modules such as warehouses, supply chains, and production planning, and will engage more roles across departments.
From an internal application to more complex business systems, Feishu CodeM is entering more real-world development scenarios, giving AI the opportunity to become a new “colleague” in R&D teams, consistently taking on tasks previously performed manually by developers.
If you're interested in learning more about Feishu CodeM’s application scenarios, feel free to click to read the original article to obtain a customized solution!