Introduction
September has brought a wave of major upgrades to office AI agents, with three mainstream products from China’s large tech firms entering direct competition: Alibaba’s Qwen Office, Tencent’s WorkBuddy, and ByteDance’s Doubao Work. Each platform rolled out new capabilities within a tight time window, expanding the boundary of what AI agents can accomplish inside real enterprise workflows.
The product update timeline shows rapid iteration. Doubao Work delivered the highest update frequency. On September 2, it launched multi-agent parallel operation and desktop automation functions. One week later, it added native local Office editing, browser recording and replay, plus seamless switching between local files and cloud storage. On September 15, Qwen Office and WorkBuddy both released important co-updates for agent collaboration and hardware compatibility.
Qwen Office released its multi-person workspace feature on September 7. Users can describe collaboration requirements using natural language, and the agent generates web applications supporting up to one hundred concurrent collaborators. The service announced that its total user base had exceeded 30 million before the end of August. On September 2, WorkBuddy opened public access to its platform, onboarding hundreds of ecosystem partners and supporting over a dozen categories of hardware including voice recorders, cameras and smart earphones.
This article compares the three office agents across four real-world job roles: marketing, content operations, product management and front-end development. It records their performance on standardized practical tasks, analyzes their strengths and weaknesses, and explores the future competitive landscape and commercial models of office agent products.
1. Core Trait Summary of the Three Agents
The three products show clear differences in work style when facing identical tasks.
- Doubao Work: Delivers complete, high-quality multi-modal outputs. It is the best fit for scenarios requiring image, video and creative asset generation.
- WorkBuddy: Boasts intuitive user experience, and performs well for users with light development requirements. Its logic and reasoning are rigorous.
- Qwen Office: Offers stable and reliable basic office automation. Its performance is relatively balanced without obvious standout advantages or critical flaws.
2. Task-by-task Benchmark Across Four Job Roles
2.1 Marketing: Qwen Office is Prudent, WorkBuddy Thorough, Doubao Work Efficient
Competitor research is a classic marketing workflow. All three agents can collect background information and draft preliminary reports, yet their work styles diverge noticeably.
Qwen Office acts like a cautious administrative assistant. It first confirms document format and benchmark dimensions before starting analysis. Its web research output is concise, listing competitor names but offering limited depth for each analysis dimension.
WorkBuddy explains its research logic at the beginning of the report. It clarifies why these competitors are selected, then expands analysis with cataloged findings, market overview and competitor feature comparison. The report summarizes brand strengths and weaknesses, lists research angles and marks data sources for traceability.
Doubao Work calls its built-in multi-agent group to gather information. It delivers two versions of outputs: a web page summary and a downloadable document. It sorts competitors horizontally, compares core metrics with charts, and creates a timeline tracking competitor dynamic updates.
The second marketing task is producing creative assets: a complete marketing campaign package including main visual posters, Xiaohongshu cover images, and a 15-second promotional short video.
Doubao Work’s multi-modal ability stands out here. It designs simple storylines for the video, adds shot transitions, background music and voiceover narration. The poster layout is clean and visually consistent. Minor text misalignment exists in the static poster, but the overall asset package can be directly used for preview.
Qwen Office generates posters using common software templates. Its video output uses static poster image switching, similar to a motion graphic slideshow.
WorkBuddy’s poster workflow starts with static text animation, then product close-up shots, and finally returns to the main visual with motion effects. The visual quality is comparable to Qwen Office.
When the task extends beyond research to poster and short video generation, Doubao Work’s multi-modal capability creates a clear edge. For pure information collection and analysis, WorkBuddy remains a solid option.
2.2 Content Operations: WorkBuddy is Rigorous, Doubao Work Produces Detailed Breakdowns
Two rounds of content operation tests are conducted. The first task: collect recent industry hot topics and propose article planning directions.
WorkBuddy responds fastest, finishing the task in 11 minutes. It sorts trending topics by categories such as models, capital and policy, covering major industry events of the past week. The proposed topics are easy to use for quick supplementary research, though five suggested topics have weak operability.
Qwen Office returns text directly without generating separate documents or web pages. The content is short, grouping hot topics into three broad categories with shallow analysis.
Doubao Work uses its newly released browser automation capability. It opens web pages autonomously, reads five information sources, and outputs a heatmap-style topic clustering chart plus complete planning drafts. The topics are divided by color-coded segments and summarized into three core storylines, giving users clear directions for content creation. Its five proposed topics have the richest coverage and detail.
The second content task: analyze popular AI articles captured from platform accounts, identify trends, and propose content planning suggestions.
WorkBuddy scans 118 articles and filters 15 valid entries. It states the limitation that public account reading and interaction data cannot be obtained. It then tracks media volume, cross-platform repost scale and event continuity, synthesizing a propagation index. It supplements media preference analysis and content differentiation advice.
Qwen Office judges topic popularity by combining platform recommendation weight, reading volume and secondary discussion volume. It picks the top 10 trending articles and extracts key viewpoints. Its answer is complete, but less detailed.
Doubao Work condenses five core hotspots from 135 crawled articles and selects 18 for topic planning. It further picks 7 articles for deep decomposition, marking quoted data and estimated metrics so users know which material can be reused.
Across both rounds, Doubao Work produces the most complete deliverables, with smooth workflows from trend collection to topic decomposition. WorkBuddy’s method explanation is more rigorous. Qwen Office shows visible improvement in the second round.
2.3 Product Management: Doubao Work Delivers Better Mockups, WorkBuddy Provides Detailed Analysis
Two classic product management tasks are tested: creating a mini-program prototype for a milk tea ordering service, and building a data dashboard with diagnostic analysis.
For the mini-program prototype:
WorkBuddy completes the task in 6 minutes. Its output mimics a real mini-program, with simulated product images and a popular ranking filter. It takes the lead on feature richness.
Qwen Office spends 15 minutes. The function coverage is complete, yet the web page layout is plain and lacks visual polish.
Doubao Work’s result is closest to a commercial milk tea mini-program. Its homepage supports sliding banner switching, with well-matched placeholder images and unified visual style, achieving the best aesthetic effect among the three.
For the data dashboard and root-cause analysis:
All three agents build dashboards with charts and interaction modules. The main difference lies in analysis depth and presentation.
WorkBuddy writes detailed root-cause analysis, separating direct causes and fundamental causes, and proposes improvement priorities plus data calibration instructions.
Doubao Work’s analytical structure resembles WorkBuddy. Its highlight lies in next-month target definition: it defines the logic behind each target value in detail and attaches auxiliary monitoring metrics with summary comments.
Qwen Office’s analysis is concise. It meets basic requirements but does not extend reasoning after setting next-month targets.
WorkBuddy earns the recommendation for product management work for its thorough analysis, while Doubao Work has a clear advantage in UI prototype aesthetics.
2.4 Front-end Development: Doubao Work Achieves Higher Completion, WorkBuddy Shows Clever Design
The front-end test task is to build an SVG animation of a person riding a bicycle. The agent must handle code, graphics and physical relationships at the same time.
WorkBuddy’s design is creative: it draws a red scarf and fish in the rider’s pocket. The character moves smoothly, but the fish stays stuck to the body and the swimming animation is unnatural.
Doubao Work generates more natural animation. The rider’s visual presentation has no obvious physical collision errors. Cycling speed is split into three gears: leisure, normal and fast.
Qwen Office contains physical logic bugs. The rider fails to hold the handlebar, with both feet resting on the same side pedal, violating real-world physics.
The second front-end task builds a tech summit registration website with ticket sales, backend password management and attendee export.
Doubao Work’s finished site resembles a genuine event registration page. Ticket permissions are clearly defined, and the layout prioritizes event information before ticket selection, fully meeting core requirements.
WorkBuddy uses a broad blue-purple style. The basic functions are fully implemented.
Qwen Office uses its new multi-person workspace. It also adopts blue-purple tones and supports multi-user registration submission and backend attendee statistics.
Doubao Work wins this round. WorkBuddy delivers creative ideas but suffers from unstable implementation.
3. Future Competition Direction and Business Models of Office Agents
In recent months, large tech firms have consolidated their scattered office AI product lines. Tencent merged part of QClaw business and teams into WorkBuddy in July. ByteDance consolidated its product teams around Doubao Work in August, releasing the official version on August 3. Alibaba launched QoderWork and MuleRun alongside Qwen Office. WorkBuddy rolled out updates across iOS, Android and HarmonyOS mobile clients as early as July.
The benchmark tests show that basic capabilities of the three products converge quickly. All can search materials, generate documents, analyze data and build web pages. The gap increasingly comes from access to real workplace context.
- Doubao Work is tightly integrated with ByteDance’s internal ecosystem. Corporate group chats, documents, projects and organizational data can all become context sources for the agent.
- WorkBuddy connects Tencent’s IM, documents, email, meeting and knowledge base, extending entry points to headsets, cameras and other hardware.
- Qwen Office is deeply coupled with DingTalk, Alibaba Cloud and Alibaba’s enterprise customer system.
The next competition for office agents hinges on access to authentic workplace data and permission control to complete tasks end-to-end. Pure model capability advantages may be caught up within months, while internal data, permission systems and business process integration are much harder to replicate.
Business models will also evolve. Slide generation, PPT production and web page construction will gradually become standard baseline capabilities. The core differentiated value will shift to enterprise seats, professional workflow orchestration, private data integration and third-party tool calling. This trend squeezes the living space for independent office agent startups. General-purpose functions will become cheaper, leaving vertical industries such as legal, finance and medicine, which require specialized data and complex workflows, as the main battlefield for differentiated startups.
When building multi-agent enterprise systems, developers often need unified control over multiple model endpoints. 4sapi, an API gateway, helps manage cross-model access and credential routing in complex office agent pipelines.
Conclusion
The September batch of upgrades marks office agents moving from demo showcases into daily professional workflows. Qwen Office, WorkBuddy and Doubao Work each have distinct strengths across marketing, content operations, product prototyping and front-end development.
Doubao Work excels at multi-modal asset generation and complete end-to-end task delivery. WorkBuddy stands out for rigorous reasoning, detailed analysis and intuitive user experience. Qwen Office delivers stable and balanced basic office automation.
Beyond model performance, the long-term competition lies in ecosystem integration, workplace context access and enterprise permission systems. General capability will gradually commoditize, while vertical business scenarios and deep workflow integration become the core competitive barriers. For enterprises adopting office agents, the priority is to evaluate how well the agent connects to existing internal tools and business data, rather than only comparing isolated benchmark task results.
International access: https://4sapi.com
Domestic access: https://4sapi.cn
Top comments (0)