Alone, with a laptop, opening a Feishu chat window while calling upon the Feishu CodeM R&D agent.
On an ordinary workday in August 2026, a young employee at Contemporary Amperex Technology Co. Limited (CATL) picked up his lunch tray in the cafeteria and, before sitting down, pulled out his communication device to log into the internal social app "Queqiao Meeting."
The homepage recommendation list had switched to a warmer color scheme. AI filtered and presented 30 matching profiles, each card now featuring an additional "View Details" button at the bottom. The employee swiped through several screens, added potential matches to his favorites—completing the entire process in under a minute.
The driving force behind this update was a software engineer from CATL’s IT department. He was responsible for maintaining multiple internal systems, with Queqiao Meeting being just one of them. Collaborating with him on this iteration was an AI R&D agent named Feishu CodeM.
One Person Managing Multiple Projects Simultaneously
Launched in 2020 and initiated by CATL’s labor union, Queqiao Meeting started with a simple mission: to provide eligible employees with a trustworthy platform for matchmaking.
Consumer batteries are the company's core business. In recent years, rapid industry growth has led to a surge in employee numbers, with a disproportionately high percentage of young staff. With numerous factories, frequent shift rotations, tight work schedules, and the relatively small city size of Ningde, Fujian, young employees naturally have limited access to broader social circles. The app was created precisely to help employees connect within the company environment.
In its early days, Queqiao Meeting offered basic functionality: registration, profile filling, browsing, and expressing interest. Though user numbers were low, engagement remained steady. Some employees even found partners through the platform and reported back to the union, receiving little more than a brief word of thanks.
By 2026, both account volume and daily active usage of Queqiao Meeting had increased significantly compared to previous years. Ahead of the Qixi Festival, the union hoped to improve the accuracy of match recommendations and make the interface more romantic and contextually appropriate. Younger generations’ aesthetic preferences and habits had evolved as well, requiring updates to details such as color schemes, layout sections, and date formats.
The request eventually reached the IT department and was assigned to a specific development engineer. He wasn’t a dedicated developer for this app. His responsibilities spanned various internal systems, employee services, and several lightweight applications. Multiple IDE windows were constantly open on his desktop.
A single individual managing multiple projects, each with clear launch deadlines—this is a common reality in many large enterprise IT departments. Business demands keep coming, yet development resources remain stretched thin. Under traditional workflows, engineers would manually write code, consult documentation, and adjust styles, often forced to make trade-offs due to time constraints. In past years, Queqiao Meeting might only receive critical bug fixes during this period, while those “nice-to-have” minor improvements would be postponed until the next quarter.
This time, however, things were different.
Feishu CodeM Takes Over the “Minor Adjustments”
In early August, the engineer installed Feishu CodeM on his laptop and opened a Feishu chat window. Instead of drafting a requirements document first, he directly typed one sentence into the chat: "Change the background of the recommendation list cards to a warm color tone."
Feishu CodeM accessed the project code within its authorized scope and accurately located the relevant front-end style files. After making modifications, it replied with a screenshot showing both the code changes and the actual visual effect. With just a quick glance, the engineer could determine whether the color and layout met expectations and decide whether to accept the changes.
Then he moved on to the next item.
• Unify the date format to "Month Day".
• Reorder several fields on the personal profile page.
• Limit the AI recommendation list to a maximum of 30 entries, sorted in descending order by MatchScore.
• Re-add the "Details" button below both the AI recommendation and latest lists; keep the "Favorite" button.
Each request clearly specified the target page, field, and rule. For every change proposed by Feishu CodeM, the engineer reviewed the code himself, ran local tests, and decided whether to keep it or make corrections.
At one point, a question arose: How was the original sorting logic for the AI recommendation list defined? Why was MatchScore chosen over other metrics? Relevant discussions were scattered across months of chat history in the Queqiao Meeting requirement group, spanning nearly a year. In the past, he would have had to manually scroll through old messages or ask colleagues who participated in earlier conversations.
This time, he authorized Feishu CodeM to access messages from the designated group. Within the permitted scope, Feishu CodeM searched through nearly a year of chat records, identified key information segments, and provided contextual analysis and suggestions—all within the conversation interface. The engineer could directly see cited historical data and reasoning, enabling informed decisions on what to adopt.
To him, Feishu CodeM handled more than just code. Project codebases, technical skills, plugins, and reusable knowledge could all be accessed within the same workspace. For developers maintaining multiple related projects, having code, context, and tools unified in one place allowed the AI to truly understand the tasks at hand.
He did not delegate the final steps—testing, acceptance, code merging, and deployment—to any tool. Every page, every navigation, every recommendation result in the released version of Queqiao Meeting was personally clicked through, inspected, and confirmed by him. On the day before launch, he was still monitoring page load speeds and noticed one avatar loaded slowly, prompting him to fine-tune the caching strategy once again.
One person, one Windows laptop, one Feishu chat window, one Feishu CodeM R&D agent. A soft launch on August 12, followed by full promotion 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 "Too-Busy-to-Handle" Minor Tasks
As for how much time was saved, the engineer gave a conservative estimate. "The scope of adjustments this time wasn't huge—cutting development time in half seems accurate."
Yet when looking at individual tasks, the efficiency gain was strikingly clear. "As long as you describe a certain page or section—what content should appear, how dates should be formatted—Feishu CodeM can precisely assist with the adjustment."
In the past, modifying display formats required locating the correct component, identifying where the date formatting function was called, editing the code, running tests, and checking results—a cycle that typically took about half an hour. With Feishu CodeM, from typing "unify date format to Month Day" to seeing the correctly displayed page, the process now took only a few minutes.
The most significant change occurred in micro-optimizations that were previously often deferred. These tasks weren’t technically difficult, but shared a common obstacle: they needed to be done, yet there was never enough time.
With Feishu CodeM in place, the likelihood of completing these small refinements on the fly increased dramatically. The developer remained the same, juggling the same number of projects, but now each project had more room for meticulous polishing.
For an application serving internal employees, these details are far from trivial. Slightly smoother interactions, clearer information presentation, and more accurate recommendations ultimately shape the real user experience.
Wherever Requirements Extend, AI Steps In
The Queqiao Meeting case served as an entry point for CATL’s exploration of Feishu CodeM.
Throughout the process, Feishu CodeM did more than revise code. It immersed itself in a real corporate R&D environment, understood project contexts, took on concrete development tasks, and collaborated with engineers throughout the journey from requirement to code delivery. Engineers retained responsibility for requirement evaluation, architectural design, and code review, while the AI handled machine-executable development work.
This experience further revealed to CATL that the value of a coding agent goes beyond improving efficiency in individual coding tasks. When AI integrates into existing enterprise R&D workflows and participates dynamically alongside evolving requirements, developers can redirect more of their attention toward tasks requiring professional judgment.
In fact, CATL plans to fully adopt Feishu CodeM in the upcoming development of its prefabricated inventory management system. Compared to Queqiao Meeting, this system is much closer to core operations, covering modules such as warehousing, supply chain, and production planning—and will involve many more roles.
From internal tools to increasingly complex business systems, Feishu CodeM is stepping into more authentic R&D scenarios. AI thus gains the opportunity to become a new "colleague" in the daily workflow of development teams, continuously taking on tasks that developers used to complete one by one.
If you're also interested in applying Feishu CodeM, click the link below to get your exclusive solution!