In 2026, the focus in artificial intelligence shifted from model parameters to what AI implementation can provide. The solution proposed by ByteDance is the integration of large model capabilities into the workflows of various industries. For example, the Doubao agent was tasked by the marketing team with performing a daily task: at 9 am, it uses its skills to collect data, analyze, and create a competitor summary that is then sent to the Feishu group.
Seedance 2.5 was recently released, and Xpeng was among the first companies to confirm cooperation. This trend is supported by policy initiatives: in July, the State Council approved the Fifth Five-Year Plan for consumption expansion, promoting the concept of 'AI plus consumption,' which was detailed by eight ministries in 17 measures.
Thanks to such tools, an analyst can receive a competitor summary in their Feishu group at 9 am before they even arrive, as Doubao performs the scheduled task, summarizing categories, pricing, and strategies. The professional version of Doubao, launched at the end of June with an agent-oriented office mode, supports operation on a local computer and in a browser, as well as application and website generation, and scheduled task execution.
By June 2026, Doubao's daily token requests exceeded 180 trillion, more than ten times higher, and the number of users of AI-based applications in China reached 499 million monthly active users, with an average usage of 92.7 times per person per month. A product manager was able to ask Doubao Professional to develop a prototype of a minimalist AI office mobile interface, and a complete prototype with documentation was received within minutes, eliminating the need for multiple message exchanges between engineers and managers.
Scheduled tasks transform large models into colleagues who work at set times, integrating into existing workflows. The visual revolution is significant in larger industries. Seedance 2.5 was launched jointly with XCMG and Xpeng via Volcano Engine. In embodied AI, systems such as AgiBot, Lingchu, Qiongche, Weifen, and Xspark AI connected to Seedance 2.5 to generate interaction data compliant with physical laws for cross-domain training, thereby replacing scenario-based simulation.
Yanjun Wu, Director of the Institute of Software, Chinese Academy of Sciences, noted that one of the main problems is the shortage of high-quality training data: human collection is unrealistic, extreme conditions are inaccessible to people, and scenario modeling is tedious; synthetic data from Seedance can significantly reduce costs and development cycles.
Mirek Dusek from the World Economic Forum noted that the discussion of enterprise AI is now focused on creating real business value. The Mercedes-Benz CLA electric vehicle integrates Doubao: for instance, a user can request to find a Cantonese restaurant with charging and air conditioning reduction, and the system will handle the intent. Volcano Engine President Tan Dai established the evaluation criterion—token economic efficiency. He stated that the price of a single token is rising, but the value created by it is rising even faster. A semiconductor manufacturing client delegated the design of an RTL chip to Doubao 2.1 Pro, which produced 1300 lines of executable code in 18 hours; this represents production-level code that passes tests and goes into operation. Doubao 2.1 Pro outperforms Opus 4.6 in coding and agent tasks, creating positive feedback for Kingsoft Office and Chopstick.