Baidu Bets on Context as AI Office Competition Moves to Enterprise and Overseas

For Baidu, the edge comes from Wenku and Netdisk, both operated for more than a decade. Wenku has accumulated public knowledge and user-uploaded documents, while Netdisk holds personal and enterprise data including files, images and videos. These assets were once archives, but in the AI office era they can be retrieved and generated, Lei Feng News reported. In April last year, Baidu Wenku and Netdisk jointly launched GenFlow 1.0, entering task scenarios such as writing papers, making PPTs and generating industry reports as a general AI Agent. In August this year, GenFlow was officially named "Kuku AI" in Chinese and launched an independent app.

Data being called does not automatically make it context. Zhong Hao, head of PSIG domestic business, said the key is to move data "from accumulation to use, from fragmented to structured" so AI can absorb and reuse it. The team has focused on deepening content understanding and letting AI penetrate. For example, Netdisk data is raw material; only when understood, structured and recalled in tasks does it become context.

Work context is complex and not ready-made. Communication occurs in tools such as Feishu and DingTalk, information comes from search, and business sits in specialized software and systems. There are natural "context walls" between systems. Wang said Kuku AI has opened connector authorization, so content from Feishu, DingTalk and other platforms can be called by Kuku AI after user authorization. Baidu Netdisk is also activating its content assets to become a source of users' work context.

Lei Feng News learned that for Word, PDF, PPT, images and videos in Netdisk, Baidu has been doing multimodal content understanding and file format parsing, then identifying entities, concepts and relationships to structure scattered content so agents can call it. Tang Haihao, head of PSIG basic R&D and B-end business, said parsing and structuring multimodal data is a bigger challenge than plain documents, and it is where Netdisk hopes to differentiate.

To become a true productivity tool, AI office products need enterprise context, not just personal context. Wang said the difference between individual and enterprise is not just user scale. Many work scenarios require organizational collaboration, so context's value to an organization lies in collaboration, knowledge accumulation and experience reuse. Enterprise context requires AI to understand not only "me" but also "the organization," participating in collaboration and preserving knowledge and experience for others and other agents to reuse.

This idea runs through Kuku AI enterprise edition. Tang said closed B-end clients mainly value three things: directly solving problems, security, and accumulating experience. In finance, especially investment research, customers buy when results are good; in education, teachers who make PPTs buy directly. If every employee uses a separate agent, enterprises worry about data security, so bosses value enterprise-level security. Some bosses want to retain top employees' experience as company assets.

Enterprise context is also becoming a new consideration in B-end procurement of AI office products. Lei Feng News learned that Kuku AI enterprise edition has just launched and 3,000 enterprises are already trialing it, with faster adoption among small and medium-sized enterprises. PSIG chose this direction because large enterprises have long processes and slow decisions, making restructuring difficult, while SMEs transform faster. Multimodal capabilities also helped it win customers whose work centers on images and videos.

Wang said Kuku AI has run through both To C and To B scenarios, and the two can reinforce each other. To B scenarios accumulate expert skills that can feed To C; To C office demand data helps improve agent data and general intelligent agent capabilities, which then serve more demanding To B customers, forming a data flywheel.

Baidu is also looking overseas. Last year, Baidu Wenku and Netdisk launched Oreate AI, a one-stop overseas AI office platform, whose overseas users exceeded 10 million within a year. Oreate AI has now been renewed as Kooko, the overseas version of Kuku AI. Wen Siying, head of PSIG overseas business, said the overseas market is still in an early stage of adoption and the "cake" is still growing; the priority is to capture user mindshare.

Overseas, AI office products face localization tests. Users in different countries have different software environments, work habits and content preferences. For example, demand for specific templates has grown quickly in South Korea; AI penetration in Latin America and the Middle East differs from the United States; and even for the same PPT, markets differ in information density, design and color preferences. Lei Feng News reported that going overseas is not just moving a product to another market, but placing AI in another work environment so it can understand local ways of working—arguably rebuilding "work context."

Editor's Summary

Baidu is positioning context as a central battleground for AI office, using Wenku and Netdisk content assets and Kuku AI to connect personal, enterprise and overseas work scenarios. Its enterprise edition has drawn 3,000 trial companies, while overseas Oreate AI became Kooko after surpassing 10 million users. The outcome will depend on how well Baidu structures multimodal data, breaks context walls and adapts to local work habits.