On August 25, ByteDance's Doubao officially unveiled its Agent product tailored for enterprise office environments, named Doubao Work.

Positioned as an enterprise-grade Agent solution for productivity scenarios, Doubao Work is designed to autonomously break down user objectives, invoke tools, and drive complex workflows to completion. According to the official announcement, the desktop version of Doubao Work is now available for download via the official website, and users can also access it directly through the latest version of the Doubao desktop application. After downloading, logging in, or upgrading, users can claim a free 30-day subscription trial.

This move adds another heavyweight contender to the office Agent battle already ignited by Tencent, Alibaba, and Baidu, intensifying the competitive landscape.

Capabilities: From Content Generation to Computer Control

The foundational capabilities of Doubao Work span everyday office tasks such as drafting proposals, compiling meeting minutes, preparing reports, conducting research summaries, and performing spreadsheet analysis.

In terms of content generation, the official description indicates that Doubao Work supports the creation and editing of various formats, including documents, spreadsheets, presentations, images, videos, web pages, and applications.

For editing, Doubao Work offers a collaborative feature described as "point-and-edit," allowing users to select any segment within a document, spreadsheet, presentation, or webpage for targeted modifications, eliminating the need to regenerate entire content pieces.

In more complex task scenarios, Doubao Work can directly operate browsers and computers upon user authorization, processing data across software platforms, collecting information, filling out forms, and comparing materials. For long-duration tasks, it can leverage cloud-based computers to execute continuously, ensuring progress is maintained even if the local device is powered off. Additionally, users can remotely control their computers via mobile devices, enabling them to continue unfinished work while on the move.

Regarding multi-Agent collaboration, users can access tools specialized in areas such as data analysis, content planning, scientific research, and finance. They can also assemble multiple professional Agents into "work squads" for division of labor—for instance, coordinating data, research, and design Agents to collaboratively produce a data research report.

Integration with Feishu: Enterprise Context as the Core Differentiator

The key distinction of Doubao Work from standard Agent tools lies in its deep integration with Feishu (Lark).

When logging in with a Feishu account, Doubao Work can, within user permission boundaries, access information accumulated within Feishu, including chat records, documents, meeting minutes, and schedules. This enables task completion based on real enterprise context, rather than relying solely on a single user command.

Officially, content and knowledge co-created by users and Doubao Work will also be continuously fed back into Feishu, becoming enterprise knowledge assets that can be re-edited, shared, and reused. Following the launch of Doubao Work, the existing Doubao Enterprise Edition entry point within Feishu has been upgraded to reflect the new product.

On the security front, Doubao Work inherits Feishu's existing permission system, ensuring users can only access data and utilize tools within their own authorization scope. Personal and enterprise data are kept isolated, with protections spanning device access, permission settings, quota controls, data encryption, and operational auditing. Currently, Doubao Work has become one of the first office Agents in China to pass both the office intelligence capability and cloud-based benchmark tests.

The Office Agent Battle Enters a 'Four-Power' Showdown

Competition among Alibaba's Qianwen Office, Tencent's WorkBuddy, and Baidu's "Kuku AI" has already reached a fever pitch. The launch of Doubao Work marks ByteDance's formal entry into this fiercely contested arena.

On August 14, Alibaba's Qianwen Office announced the integration of two external models, Z.AI's GLM-5.3 and DeepSeek V4 Pro. On the same day, Baidu Wenku's general-purpose Agent officially adopted the Chinese name "Kuku AI" and launched a standalone desktop client, also incorporating external models from DeepSeek and Z.AI. Meanwhile, Tencent's WorkBuddy has expanded its built-in model pool to include series from Hunyuan Hy3, Z.AI, MiniMax, Kimi, and DeepSeek, among others.

ByteDance previously introduced TRAE Work, which features built-in models including ByteDance's Seed series, Z.AI GLM, DeepSeek, and Qianwen. The release of Doubao Work represents ByteDance's further strategic push in the enterprise office Agent sector.

From Tencent and ByteDance to Alibaba and Baidu, leading domestic tech firms are increasingly converging on a model aggregation approach for office Agents. According to analysis, "In the past, large model products competed on 'whose model is stronger.' At the office Agent stage, the competitive logic has shifted: first, place different models into a single workspace to encourage more users to entrust their real work to it."

Real workflow data is emerging as the core bargaining chip in this competition. Wang Ying, Vice President of Baidu and President of the Personal Super Intelligence Group, stated, "As public data is progressively learned by models, the new knowledge, experience, and ideas generated by humans will become increasingly important as AI continues to improve."

However, model aggregation is just the first step. The next critical question is routing—determining which model should handle a specific task. Large model deployment faces a trade-off among effectiveness, speed, and cost. "Once office Agents enter large-scale use, even minor differences in per-call costs—whether a few cents or a few dimes—will be amplified significantly with increasing task volumes."

Goldman Sachs Research: Agents Enter the Execution Era, AI Competition Shifts to Workflows

This assessment aligns with Goldman Sachs' recent field research on the Silicon Valley AI supply chain.

Goldman Sachs believes AI commercialization is transitioning from "per-seat subscriptions" to charging based on consumption volume, transaction volume, and outcomes. Simultaneously, Agents are evolving from assistive tools into workflow executors, with industry value shifting from the models themselves to proprietary data, business context, and domain expertise.

This implies that AI industry competition is moving from "whose model is superior" to "who can truly master workflows." Model capabilities remain important, but the ability to integrate into enterprise production environments, understand business context, and reliably execute tasks will become a more critical competitive moat.

Goldman Sachs projects that computing power demand could grow by approximately 24 times over the next five years.