Doubao wildly adds product features, the AI battlefield remains unclear

Author | Liu Baodan

Editor | Huang Yu

The AI application chase is still ongoing. Since being surpassed by DeepSeek at the beginning of the year, Doubao has accelerated its product iteration pace.

On June 17th, Wall Street Insight learned that the Doubao desktop version has fully launched the AI podcast feature. Users can click on "AI Podcast," upload a PDF or webpage link, and generate a podcast program in a two-person dialogue format with one click. Currently, the Doubao App has also started a small traffic test and will be fully launched soon.

Unlike traditional AI podcasts, which are often not conversational, sound mechanical, and lack interaction, this capability is based on the voice podcast model developed by Doubao's large model team, allowing for a seamless transformation from text creation to a two-person dialogue podcast in just seconds, achieving a "low-cost, high-efficiency, strong interaction" creative experience.

This is just one detail of Doubao's product iteration. According to Wall Street Insight's observations, since March, Doubao has significantly accelerated its product iteration speed, first upgrading the "AI Programming" feature launched in January, followed by testing the new deep thinking feature, and subsequently, Doubao has seen multiple updates in its image and video capabilities.

Last week, Doubao's large model 1.6 was officially released, further upgrading Doubao's video capabilities, which can now generate seamless multi-camera switch videos as well as video creations in various styles such as realistic, animation, and film. As ByteDance's main AI product, Doubao continues to enhance the usability of its products.

Behind the frequent product launches, Doubao is attempting to attract user attention through new features to maintain user activity levels.

This is not an easy task. At the beginning of the year, DeepSeek exploded in popularity overnight, bringing in a large number of new users. QuestMobile data shows that by the end of February 2025, the user scale of AI native apps reached 240 million, a month-on-month increase of 115 million, which is nearly a doubling growth. However, after the peak, the industry inevitably faced a decline in users.

According to QuestMobile data, by the end of May 2025, the active user numbers of applications such as DeepSeek, Yuanbao, Kimi, Wen Xiaoyan, and Nano AI generally declined. For example, Yuanbao's monthly active users dropped from 41.64 million in March to 22.56 million in May, losing nearly half of its users.

In contrast, Doubao has become one of the few AI applications with continuous user growth, with its active user count reaching 131 million in May, an increase of about 15 million over two months, further closing the gap with DeepSeek.

Currently, the growth model of AI products has changed. DeepSeek not only brought a major reshuffle of native apps but also made the industry realize that the iteration of AI technology remains the core driving factor for products. In other words, a leap in AI technology can directly crush user growth brought about by market spending.

This is indeed the case; apart from Yuanbao and Quark, which conducted large-scale investments in February and March respectively, the investment efforts of Doubao, Kimi, and others have significantly weakened. AppGrowing data shows that Yuanbao's investment amount in March was approximately 1.386 billion yuan (estimated), and it has since maintained a level of 600 million to 700 million yuan, while Doubao's monthly investment is less than 100 million yuan An insider from an AI unicorn company focused on C-end applications told Wall Street News that the company's current focus is on researching the next-generation model. As for product iteration, it largely depends on the company's bets on the technological path of the next-generation model, which is the core capability that determines AI products.

In fact, whether it is AI dialogue products, AI assistants, or AI browsers, they may only be intermediate forms of killer AI products. For Doubao, whether it is AI podcasts, deep thinking models, or video generation capabilities, they essentially remain optimizations of single-point functions. As user expectations for AI shift from "tool assistance" to "task solving," Agents are becoming the new competitive focus.

Currently, major companies like ByteDance, Baidu, and Alibaba have entered the Agent space. On April 18, ByteDance's Manus product "Button Space" began internal testing, focusing on the ability of Agents to solve work scenarios, allowing users to call applications like Amap, Feishu, and Feichangzhun through MCP to help solve practical problems.

Baidu has also started experimenting with Agent products. On April 25, Baidu officially launched the multi-intelligent collaboration app Xinxiang. Baidu founder Robin Li stated that future AI applications should not only answer questions but also complete tasks. Alibaba has also added Agents to Quark, and on June 12, Quark released its first Agent capable of generating complete volunteer reports.

For Doubao, the intensive product iteration feels more like a "defensive innovation," maintaining user activity through frequent updates while also reserving scenario data and technical validation for the upcoming Agent ecosystem battle. The aforementioned insider from the AI unicorn company stated that "Button Space" is still in its initial stage, and future integration into Doubao as a unified entry point cannot be ruled out.

As the window for technological breakthroughs shortens, single-point functions can no longer build a moat. Whether Doubao can leverage ByteDance's traffic and scenario advantages to seize the initiative in the Agent collaboration network will be a key battle in its transition from "user growth" to "ecological dominance."

The second half of the AI battle remains undecided