After the horse race, ByteDance calls a retreat.
On the afternoon of August 24, ByteDance completed another round of AI business integration. According to "Intelligent Emergence", a media outlet under 36Kr, the TRAE and Coze teams have been fully merged into the Doubao system. Relevant capabilities of TRAE Work and Coze will be further integrated with Doubao's work scenarios, and TRAE IDE and CLI will remain in operation but be adjusted as programming product lines under the Doubao brand. After the adjustment, relevant product and operation teams will report uniformly to Zhao Qi, head of Doubao products. ByteDance also plans to launch an independent AI office product called "Doubao Work".
It has only been 25 days since the last round of adjustment. On July 30, ByteDance announced the integration of the Feishu product team and the Doubao product team, with Zhao Qi taking overall charge; the Feishu GTM team was integrated with Volcano Engine to establish a new To B GTM organization named "Creativity Service Platform". The Feishu brand and existing products will continue to be retained, but the originally relatively complete organizational system has been split: product R&D is incorporated into Doubao, and the marketing, sales and customer service systems are incorporated into Volcano Engine.
At ByteDance's mid-year all-staff meeting on August 6, Liang Rubo gave a clearer strategic explanation. He summarized the company's business principles as "high priority, thick backbone, long-term optimization". The current core businesses are focused on AI, information platforms and transaction services, and he proposed that Doubao is expected to become a "backbone business" similar to Douyin in the future. According to Liang Rubo's explanation, the so-called "thick backbone" means that the business itself is sufficiently large and important, and can drive other closely related businesses at the same time.
Several consecutive adjustments within a month have gradually outlined the new structure of ByteDance's AI business. Feishu, TRAE and Coze, which were originally scattered in different organizations, are now rearranged around Doubao one after another; Volcano Engine takes over the enterprise marketing, sales and customer service functions, while Doubao gains product and R&D resources. The change of organizational structure has gone beyond the scope of ordinary product line adjustment. ByteDance is reconfiguring the business assets accumulated over the years in accordance with the new strategic priority.
1
After Doubao Becomes the Backbone Business
The reason why Feishu, TRAE and Coze were successively incorporated into the Doubao system in the same summer is first related to ByteDance's judgment on AI productivity. Doubao has built a considerable C-end user base. The problem ByteDance is facing now has gradually shifted from expanding the AI user entry to figuring out how to bring AI into office, programming and enterprise production processes, so that users can use Doubao to complete more complex and higher-frequency work.
The enterprise market has given some clear feedback. According to a report from 36Kr, Liang Rubo mentioned at the August all-staff meeting that more than 90% of Feishu's new customers have purchased Feishu AI products synchronously. He also judged that the development speed of AI in productivity scenarios is faster than expected, and the importance of To B business has increased accordingly. 36Kr previously also reported that since 2026, the proportion of Feishu's new customers who purchase AI capabilities synchronously has reached around 90%.
The data has changed Feishu's strategic value within ByteDance. In the past few years, Feishu developed as an independent office product, competing with DingTalk and WeCom, with its business goals mainly focused on the collaborative office market. After AI entered the core strategy, Feishu's years of accumulated enterprise customers, organizational relationships, document systems and collaborative scenarios have gained a new purpose: if Doubao wants to enter the real enterprise workflow, it needs to use the product environment that Feishu has already built.
The reference frame for the company to evaluate Feishu has also changed accordingly. Simply judging whether Feishu can independently grow into a larger SaaS business can no longer cover all its value. The company also needs to calculate how many enterprise customers, work scenarios and commercialization capabilities Feishu can bring to Doubao and Volcano Engine. The organizational adjustment on July 30 finally gave the answer: the Feishu product team was incorporated into Doubao, and the GTM team was incorporated into Volcano Engine. A originally complete business was split into two parts according to the new strategic needs.
A similar logic also applies to TRAE and Coze. TRAE originally focused on AI programming, and Coze mainly provided an agent development platform. Both teams had carried out independent explorations around Agent before. By 2026, productivity Agent has become an important direction for Doubao, and the product boundaries between programming, agent development and office scenarios are getting closer and closer. Continuing to maintain multiple independent organizations will increase the cost of repeated R&D and internal collaboration. The 36Kr report also clearly mentioned that the experience accumulated by TRAE and Coze before will be reused in Doubao.
The integration on August 24 is more appropriately understood as a resource consolidation. ByteDance has determined that AI productivity is an important task that Doubao needs to undertake, and the teams that originally undertook the exploration task no longer need to be configured separately. The retention of TRAE IDE and CLI shows that the company has not abandoned the verified products just because of the organizational merger; the change mainly lies in the ownership relationship, and the programming products will serve Doubao's larger AI product system in the future.
Doubao thus has truly obtained the organizational significance of a "backbone business". To judge the strategic focus of a large company, budgets and executive statements can only provide part of the information, and more specific signals come from reporting relationships and resource flows. Feishu has been operating for many years, and TRAE and Coze also have clear product positioning respectively. ByteDance is still willing to break the original organizational boundaries and hand over the mature teams to Doubao, which shows that AI has affected the internal resource allocation rules of the company.
2
A Relatively Stable Organizational Habit
From a long-term perspective, Doubao's integration idea in 2026 is not isolated. Many of ByteDance's major adjustments in the past have similar characteristics: business boundaries will be redefined with strategic changes, and once-important products do not have permanently fixed organizational positions.
In November 2021, after Liang Rubo took over as CEO, he sent an all-staff email announcing that ByteDance had established six business segments: Douyin, Dali Education, Feishu, Volcano Engine, Nuverse and TikTok. One very representative change is that Toutiao, Xigua Video, Search, Encyclopedia and domestic vertical service businesses were all merged into Douyin, which took overall charge of domestic information and service businesses. Toutiao once long assumed ByteDance's core user entry and brand recognition. After short videos became the new traffic center, Toutiao also lost its independent organizational position.
Feishu then went through another round of changes. In the 2021 adjustment, Feishu, EE and EA were merged into an independent Feishu business segment, whose head reported directly to Liang Rubo; in July 2026, the Feishu product team was incorporated into Doubao, and the GTM system was incorporated into Volcano Engine. Five years ago, the company hoped to concentrate relevant businesses in Feishu; five years later, AI has changed the enterprise software market and ByteDance's own strategic priorities, and the original segment has been split again.
Nuverse's experience is more drastic. In November 2023, ByteDance drastically downsized its gaming business, seeking to spin off launched products with good performance, shutting down most unlaunched projects, and only retaining a small number of innovative projects and related technical explorations. In March 2024, ByteDance released another internal letter, returning the gaming business to an "incubation state", readjusting the leadership and organizational structure, and continuing to retain UGC, distribution and part of R&D capabilities. The gaming business did not continue to expand along the original scale, nor did it disappear completely from the company. The company reduced its scale according to the new input-output judgment, and then decided which parts are worth retaining.
The adjustment of the education business is more directly affected by policy factors. After the "double reduction" policy in 2021 changed the environment of the online education industry, Dali Education quickly downsized, and businesses such as Guagualong and Qingbei Online School experienced layoffs and adjustments. After fundamental changes took place in the external market, the organization formed by ByteDance's previous investment could not maintain its original scale solely by virtue of historical investment.
Similar cases illustrate a relatively stable organizational habit of ByteDance. What makes ByteDance special is not that it "dares to lay off employees" or "dares to shut down businesses" — large technology companies all have a large number of failed projects and organizational adjustments. What is more noteworthy is the frequency with which ByteDance adjusts business boundaries, and the fact that the company rarely maintains a corresponding independent organizational status for a long time just because a business was once important.
While business identities can change, capabilities that have been proven valuable often find new places to go. After Feishu was split, the product team was incorporated into Doubao, and the sales and service system was incorporated into Volcano Engine; after the TRAE team was merged into Doubao, IDE and CLI continued to develop as programming product lines. The change of organizational name does not mean that all the previous accumulation is zeroed out, and the company will continue to use products and capabilities that can fit into the new strategy.
Therefore, summarizing ByteDance with "cutting off businesses" is likely to miss the key part. What ByteDance often does is more like re-splitting assets: after the original business is split, the company judges which products, technologies, talents and customer resources are worth continuing to allocate, and then puts them into the new business structure. When a project no longer needs to exist independently, it does not mean that all resources within the project lose their value at the same time.
This approach can reduce a type of organizational cost common in large companies. The longer an enterprise develops, the easier it is for products, teams, leaders and budgets to form a fixed relationship. Even if the strategic value of a business has declined, the original organization may continue to exist due to historical inertia. ByteDance is willing to frequently break the formed organizational boundaries, which allows resources to move to new directions faster, but it also inevitably sacrifices part of the team stability.
3
Serving the Same Company Center
When the outside world talks about ByteDance's organizational culture, they often mention "horse racing". When the direction is not yet clear, allowing multiple teams to work in parallel in the same field can increase the probability of finding an effective product route, and also reduce the risk of the company betting on a single path prematurely. A large number of ByteDance's early products grew up in a similar environment.
The AI office sector in 2026 provides a very complete sample. At the beginning of the year, Doubao, Feishu, TRAE and Coze entered the productivity market from the dimensions of general assistant, collaborative office, AI programming and Agent development respectively, and each team had relatively independent exploration space. By the middle of the year, market demand and the company's product direction gradually converged. The Feishu product team, TRAE and Coze were successively incorporated into Doubao, and Feishu GTM was incorporated into Volcano Engine. The original multiple routes were reorganized into a more centralized structure.
Horse racing and merging actually correspond to two different stages. When the market answer is not clear, the company is willing to bear a certain degree of internal repeated investment, and use more product experiments to obtain information; when the company has confirmed the priority direction, continuing to maintain similar teams will increase resource consumption, and the organization will turn to centralization. There is no fixed sequence of dispersion and concentration, and the key depends on whether ByteDance's judgment on market opportunities has changed.
Judging from the information disclosed at the August all-staff meeting, ByteDance obviously believes that AI has entered a stage where resources need to be concentrated. Liang Rubo included AI in the three core businesses, and clearly proposed that Doubao is expected to become a "backbone". Subsequently, TRAE and Coze were fully incorporated into Doubao. The interval between strategic statements and organizational actions is very short. Once the company determines the priority, the department reporting relationship will change quickly accordingly.
The so-called "virtual corporate self" can also get a more specific explanation based on this. There seems to always be a resource allocation center within ByteDance that is higher than a single product, and the company will re-determine priorities according to user scale, market opportunities, technological changes and commercial returns. Toutiao was once at the core of domestic information distribution, and later its related businesses were incorporated into Douyin; Feishu was once a first-level business segment, and after AI became the core strategy, Feishu was reconstructed into Doubao and Volcano Engine. The status a product achieved in the past cannot be automatically converted into its future organizational status.
"Making full use of everything for our own purposes" is very suitable to describe ByteDance's way of dealing with business assets. The "our own" here refers to a "corporate overall interest center" that is higher than specific products and business departments, which represents ByteDance's unified judgment on growth, resources and strategic direction at a specific stage.
Under the same strategic goal, ByteDance maintains relatively low integrity for immature businesses, and resource reallocation occurs more frequently. As long as the company judges that the new combination can better serve the current strategy, product names, department boundaries and reporting relationships can all be adjusted. Businesses serve the company's strategic goals, and the company rarely restricts the flow of resources in order to maintain the original organizational form of a certain business.
ByteDance's horse racing culture thus has its other half. The outside world can easily see multiple teams working on similar products at the same time, but rarely notice what happens after the horse racing ends. What really determines the efficiency of horse racing is not only whether multiple projects can be run out at the same time, but also whether the company has the ability to stop repeated investment after the results come out, and quickly put the formed resources into the winning direction.
A strong resource allocation center will also generate costs. Once the judgment of the main track is wrong, the faster the concentration speed, the larger the scale of misallocated resources may be; frequent changes of organizational affiliation will also reduce the team's expectation of long-term business boundaries. The gaming industry can particularly illustrate this point. Nuverse has experienced large-scale investment, strategic downsizing and re-incubation, and the business must ultimately follow the industrial law of long game R&D cycle and low success rate. Organizational adjustment itself cannot solve the difficulties brought by product laws.
Therefore, ByteDance's truly worthy of study organizational capabilities can be embodied in two links. In the early stage, multiple teams are allowed to find answers, and after the market direction gradually becomes clear, the company has sufficient power to reconfigure the original businesses. The changes that took place between Feishu, TRAE, Coze and Doubao in the summer of 2026 just connect the two links together.
ByteDance will certainly allow specific business organizations to operate around certain mature businesses for a long time, but the more common situation is that the business constantly adjusts its position as the company's strategy changes. Horse racing and merging are both means of resource allocation. What remains truly stable for a long time is that ByteDance always keeps the power to redefine business value in its own hands.
This article is from the WeChat official account "Xiutai", author: Shi Can, published with authorization from 36Kr.