ByteDance is reportedly changing the organization behind Doubao, its fast-growing AI app, as it shifts resources across chat, monetization, workplace tools, and model post-training. The changes come after the app surpassed 200 million daily active users, according to TechNode’s September 23 report, though that later user figure was not independently verified in the sources reviewed.
TechNode reported that the company is reducing Doubao’s general Session team by about half, citing a LatePost report. LatePost puts the group at about fifty staff members whose remit included strategy and evaluation for Doubao’s chat product. Some affected staff may transfer to Doubao commercialization and Feishu teams, while others may be laid off. ByteDance has not publicly confirmed the reported restructuring in the material reviewed for this article.
The story belongs to a longer reorganization of ByteDance’s AI business. LatePost reports that a Feishu-product/Doubao-product-team integration occurred in July 2026, an evolution also discussed in EastFrontier’s August Doubao and Feishu coverage. The latest account points to a more specific adjustment around the general-chat experience.
A smaller team around the general chat experience
LatePost, in a report republished by X-TechCon, says the general Session group had responsibility for the strategy and evaluation of Doubao’s chat capabilities. TechNode summarizes the reported change as a reduction of roughly half. The source language matters. This is an account of internal decisions, not an official layoff notice, and the reported numbers should be treated as approximate.
The same reporting says ByteDance is trimming the Product Posttrain-Chat team, which works on adapting Seed base models for the chat product. LatePost describes that group as part of ByteDance’s large-model department and says it works alongside the Doubao product team. Some employees have reportedly moved to Horizon RL and Product Posttrain-Work.
That arrangement offers a glimpse of how a large AI product changes as it matures. At an early stage, a chat application may need a dedicated team to decide what capabilities to build and how to measure them. Later, the product organization may want to consolidate the supporting models, reduce routing complexity, and move people toward parts of the business where demand is growing faster.
LatePost says the post-training group merged several smaller models used by Doubao chat into one around June, simplifying routing and reducing routing errors. The report does not provide model names, benchmarks, or an official explanation from ByteDance. It should therefore be read as a reported internal engineering move, not as a confirmed model release or proof of a particular quality improvement.
The 200 million-user milestone needs context
TechNode says the changes followed Doubao surpassing 200 million daily active users, while also noting that ByteDance continues to optimize the chat experience rather than abandoning it. The reporting does not supply the methodology or measurement date behind the 200 million figure, so it should remain attributed to TechNode and LatePost.
There is an earlier independent point of reference. Reuters reported in February that Doubao had exceeded 100 million daily active users on February 16, citing data from AICPB.com, a tracker of Chinese AI chatbots. That report established a 100 million-user snapshot months before the current 200 million claim. It does not independently prove the later figure, but it helps explain why Doubao has become large enough for changes in its internal organization to matter.
The scale also explains why ByteDance faces a different set of trade-offs from a smaller AI startup. Once an app serves tens or hundreds of millions of daily users, even minor changes to model routing, response quality, safety review, or interface design can create a substantial operational burden. A leaner chat team may make sense if the product has become more standardized, but it can also increase the importance of shared model and evaluation systems elsewhere in the organization.
The current sources describe transfers toward commercialization, Feishu, Horizon RL, and Product Posttrain-Work. They do not provide a public product plan for consumer phone features or a wider agent release. That makes the restructuring distinct from EastFrontier’s separate coverage of Doubao’s phone-assistant beta.
The limits of the reported restructuring
The reported personnel moves invite an easy but unsupported conclusion: that ByteDance is retreating from chat. TechNode explicitly says the company is still optimizing the chat experience. The sources instead point to a narrower possibility. Dialogue-related work may be receiving less standalone attention as ByteDance tries to integrate chat into a larger product and model stack.
The details also remain incomplete. The reporting does not establish how many people will leave, how many will transfer, the final size of either team, or which executives approved the changes. It does not disclose revenue, profitability, user retention, or a timetable for product changes. Names of internal leaders appear only in Chinese in the LatePost account and are not accompanied by verified English transliterations, so they are not necessary to understand the development.
One point that should not be lost is the relationship between product scale and organizational design. A chat app can become popular because it is simple to use. The company behind it may become more complex as it has to support multiple models, safety and evaluation systems, a consumer app, an enterprise suite, and new commercial services. The reported Doubao moves suggest ByteDance is trying to redraw those boundaries.
The immediate facts are limited but meaningful. TechNode and LatePost report cuts to a general Session team and a Product Posttrain-Chat group, with possible transfers to commercialization, Feishu, Horizon RL, and work-focused post-training. The reports also place those changes after Doubao’s reported 200 million daily active-user milestone. Until ByteDance comments publicly, the restructuring should be treated as reported internal reallocation, not as a definitive statement about the company’s wider AI strategy.