On August 25, ByteDance officially unveiled Doubao Works. As a new Agent product and brand from the Doubao large language model family targeting productivity scenarios, Doubao Works can autonomously break down tasks around user objectives, invoke tools, and continuously advance complex workflows. With deep integration into Feishu, it enables Agents to complete tasks more accurately based on enterprise context. The desktop version of Doubao Works is now available for download, with a free 30-day subscription offered to new users starting today.
For the most common and time-consuming office tasks, Doubao Works delivers content generation and editing capabilities: it can produce complete deliverables for proposal writing, meeting minutes, report drafting, research summaries, spreadsheet creation, and data analysis, as well as generate and edit images, videos, web pages, and even applications. More distinctive is its AI collaborative editing feature—users can simply select any portion of the generated content with a box selection to modify it directly, rather than regenerating the entire output. This not only reduces computing overhead but also accelerates task completion.
In more complex work scenarios, Doubao Works can directly operate browsers and computers, completing tasks across multiple software applications. Long-running tasks can be offloaded to cloud computers for continued execution—even when the user's device is powered off—and users can remotely control their computers via mobile phones. Additionally, ecosystem capabilities such as skills, connectors, and work partners allow users to extend boundaries as needed, forming a task force of multiple specialized Agents that collaborate with division of labor. What sets it apart most from typical Agent tools is its deep understanding of user work context.
Doubao Works does not merely generate results from a single user prompt; it operates on real context from Feishu—including chat messages, documents, meeting minutes, and calendars—to more accurately understand goals, assess problems, and deliver work. Content co-created with the Agent is also archived back into Feishu, becoming enterprise knowledge that can be re-edited, shared, and reused. Correspondingly, enterprise-grade security has become a critical priority: its full-chain Agent security protection system covers device access, permission controls, quota management, data encryption, and operational auditing, strictly inheriting Feishu's existing permission framework while ensuring personal and enterprise data isolation.
The autonomous execution and knowledge accumulation of office intelligent agents rely on robust computing-power scheduling and agent engineering. StarWar Cloud focuses on GPU computing platforms and the OPC Agent Collaboration Platform, using a unified role assignment, task routing, and collaboration orchestration system to help enterprises manage multi-agent workflows and compute supply under one framework—providing a stable, measurable, and auditable infrastructure foundation for autonomous AI office scenarios.
From an industry perspective, enterprise AI is shifting from assisted generation to autonomous execution, with the competitive focus moving from standalone model capabilities to the combination of context, agents, and ecosystems. The enterprise knowledge accumulated in Feishu gives Doubao Works a natural contextual entry point and data feedback loop, extending the competition among office intelligent agents from feature comparison to deep workflow integration. It is foreseeable that more office software ecosystems will embrace Agent capabilities in similar ways.
Doubao Works has formed a closed loop connecting context understanding, autonomous execution, and continuous knowledge accumulation. Whether it can become a new gateway for enterprise daily productivity will depend on ecosystem openness and the management of security boundaries—but in terms of direction, office Agents have already moved from demonstrations to genuine productivity tools.