Qianwen Office makes prompts into shared business tools

Qianwen Office (千问办公) turns a written requirement into a publishable webpage where up to 100 people can collaborate online, shifting the AI-office question from drafting content to operating shared work. According to geekpark, its multi-user workspace combines role permissions, a cloud database, an administration backend, and online publishing. A generated page can be published with one click, while members enter through a link and see functions based on their permissions.

That changes what a prompt can produce. Instead of a disposable answer, it can become a team tool that collects submitted data, tracks task status and work progress, and compiles those records through a backend. The useful test is therefore broader than presentation quality: can an AI-generated workspace make recurring coordination usable while keeping access and administration under control?

From generated page to working team system

Qianwen Office (千问办公) frames the output as more than a generated page. According to geekpark, a user can describe roles, tasks, and the data to be managed in natural language, then generate a collaboration tool for a team. The resulting webpage can be published directly and supports up to 100 people working online at the same time. That shifts the unit of output from a document for one person to a shared operating surface.

The feature is called a multi-user workspace.

Its structure matters because it includes role permissions, a cloud database, an administration backend, and online publishing, geekpark reports. Administrators and ordinary members can receive different rights over data and operations. The workspace can centrally retain submitted information, task status, and work progress. Its backend can manage those records and compile statistics, rather than leaving each participant with a separate copy of a spreadsheet or generated file.

Publishing is part of the workflow, not an afterthought. After generation, the page can be published with one click; members open a link and see functions according to their permissions. woshipm notes that Qianwen Office entered public beta in version 0.1.0 and could generate an interactive website available locally at an address similar to http://localhost:5177/index.html. The larger implication is practical: prompt-based creation can produce a tool with shared records and an operating boundary, rather than only a draft to hand off for later implementation.

The right pilot is a coordination bottleneck

A useful pilot starts with work that returns, involves several people, and changes as records arrive. Vendor intake is a strong example: according to geekpark, Qianwen Office can create a workspace for applications, document review, booth arrangements, and deposit-payment tracking at a large market event. That is a coordination queue, not a writing assignment. Each application can move the process forward or hold it up.

The test is whether the team can operate the process together.

Campaign approval offers a similar boundary. geekpark describes an influencer workspace that holds influencer lists, supports content review, and tracks publishing progress. A team can see the difference between a prompt that produces a draft and a shared tool that keeps changing work visible. The pilot has a defined group of users, recurring handoffs, and records that need updating after the initial request.

Store-opening preparation is another fit. Local managers can update preparations for multi-location launches while headquarters views progress across stores, according to geekpark. Qianwen Office lets users describe roles, tasks, and managed data in natural language, then publish the resulting collaboration tool to a team. It can also be modified through natural language when the business process changes. That makes a bounded process more revealing than a generic request for slides or copy.

The comparison should stay specific to the work at hand. woshipm characterizes QoderWork around local desktop files and coding assistance, MuleRun around cross-device task automation, and Wukong around DingTalk-connected enterprise collaboration. A pilot should ask which product matches the coordination bottleneck, then observe whether the generated workspace remains usable as people update it.

An agent needs an identity before it needs autonomy

An agent reaching internal systems needs a defined identity before it can be trusted with autonomy. The controlled-server model, described by the source, ties an agent to a domain account and requires access through the login identity of a specific employee.

That link makes delegated work attributable.

The model also separates what an agent may do from what its human sponsor may do. In a source example, an agent working for a tester could read and analyze code on a controlled server, then submit modifications, although the tester lacked permission to read the code repository. The agent's access was therefore governed through an assigned identity, rather than treated as unrestricted access to company systems.

Release remains a human decision. The model placed code reading in one permission stage and modification submission in another, while formal release required final human confirmation, according to the source. This matters because an agent can complete useful work before it receives authority to make the most consequential change.

The same distinction applies to a shared AI workspace. geekpark reports that administrators and ordinary members can receive different data and operation permissions, and that a generated webpage can be published with one click for team access through a permission-based link. A prompt may create the interface, but access rules determine who can use it and what they can change.

Messaging can trigger work from WeChat, QQ, Feishu, DingTalk, and other IM software, according to woshipm. The operational question is what identity performs the task after that message arrives.

Enterprise positioning does not settle product maturity

Qianwen Office (千问办公, QwenWork) is framed as an enterprise AI productivity platform for organizations, rather than a personal AI office tool, according to woshipm. Its August 3 public beta therefore belongs in an enterprise-software comparison. Alibaba has also assembled QoderWork, Wukong, and MuleRun within the product. Woshipm describes it as supporting desktop agents, cloud agents, and enterprise collaboration agents at the same time.

That positioning signals scope, not finished maturity.

WorkBuddy's enterprise version entered the market in June, making the public beta a visible competitive test rather than an escape from consumer-style product comparison. Qianwen Office has more than 20 connectors. WorkBuddy, by contrast, has five or six dozen external-service connectors, with more than 20 expected to be Tencent services. Connector count alone does not determine usefulness, but it affects how readily a workspace can meet existing company systems.

The model boundary is sharper. Qianwen Office supports only Alibaba large language models and defaults to Qwen3.8-Max-preview. WorkBuddy supports Tencent's Hunyuan Hy3, GLM, MiniMax, Kimi, DeepSeek, and custom models. That gives organizations a different degree of choice when model fit, cost, or internal preference varies. The woshipm author judges WorkBuddy stronger overall, especially because its model support is open. Alibaba's product can still present a coherent enterprise stack, yet enterprise positioning does not by itself answer questions about integrations, model flexibility, or collaboration depth.

Enterprise-agent workspace approaches: Qianwen Office, WorkBuddy and Nano Work

Choose the workspace by the operational problem, not the document format

• A team needs a temporary shared tool for an event, campaign, school-home process, or multi-location rollout, with forms, progress tracking, and different views for different participants. Use Qianwen Office's multi-user workspace when the workflow can be expressed as roles, tasks, and managed data. It can generate and publish a webpage for up to 100 people collaborating online simultaneously, with role permissions, a cloud database, an administration backend, and online publishing.
• The workflow requires administrators to collect submissions, monitor task status, and compile progress statistics, while ordinary participants should have narrower access. Use Qianwen Office's permissioned workspace pattern rather than a shared document. Administrators and ordinary members can receive different data and operation permissions, while submitted data, task statuses, and work progress can be centrally managed through a backend.
• A recurring process is likely to change after launch. Prefer a generated workspace if the team expects to revise the process in plain language. Qianwen Office is described as allowing users to modify a multi-user workspace through natural language after business processes change. Do not treat this as evidence of a substitute for specialized ERP, financial-system, or membership-system integration; the source material does not establish those integrations.
• The organization needs access to multiple model providers or wants to use custom models. Consider WorkBuddy rather than Qianwen Office. WorkBuddy supports Tencent's Hunyuan Hy3, GLM, MiniMax, Kimi, DeepSeek, and custom models, while Qianwen Office supports only Alibaba's large language models and uses Qwen3.8-Max-preview by default.
• An agent must act on sensitive internal systems, such as reading code or proposing changes, under delegated employee access. Use a staged permission design with human confirmation, following the internal 360 example: agents were linked to domain-account identities, code reading, modification submission, and formal release were split into separate permission stages, and humans retained final confirmation. Do not assume that every AI-office product provides these controls; the source material does not say.

Governance decides whether a workspace can stay

Governance is the test that separates an impressive generated workspace from a tool a company can keep. Procurement should establish who owns the workspace, who can export it, and what record remains when its rules change. It should also ask where data is retained, how long it stays, and which party can remove it. Connector access needs separate scrutiny: delegated permissions, identity integration, approval rules, residency, and an audit trail should be explicit before company systems are exposed.

Maintenance matters because the workflow will not stand still.

According to geekpark, users can alter a multi-user workspace through natural-language instructions when business processes change. That promise makes change control a purchasing question, not an implementation detail. Buyers should determine who may issue those instructions, how a changed rule is reviewed, and how the prior version can be recovered. They also need clarity on responsibility when generated business logic no longer matches an approved process.

The claim that Qianwen Office is the world's first Agent product for enterprise scenarios needs market context. woshipm cites an Analysys report saying 17 major domestic desktop AI-native office-agent platforms drew more than 60 million monthly visits in June 2026; Tencent WorkBuddy led with 20.97 million visits, while ByteDance's TRAE IDE domestic version and Alibaba's QoderWork followed. On August 3, Alibaba integrated QoderWork, MuleRun and Wukong into Qianwen Office and began public beta, according to the cited report. Competition also includes 360's Nano Work, launched on July 28, with about 20,000 preinstalled cloud dependencies, according to the cited report. A "first" label therefore matters less than governable access and durable maintenance.

Start with a bounded coordination workflow where the cost of a mistake is visible: vendor intake, campaign approval, or project-status collection. Then test whether the agent has an identity tied to a specific employee's domain account, as 360 required, rather than a shared or opaque credential.

Permission design is the harder test. At 360, an agent could read and analyze code and submit modifications on a controlled server for a tester who lacked repository access; code reading, modification submission, and formal release were separated, with people keeping final confirmation. Ask to see equivalent stages for the workflow you plan to automate.

Also inspect the execution environment. According to its reporting, Nano Work's cloud office includes about 20,000 preinstalled dependencies. That breadth may make temporary tools practical, but it raises a deployment question: can the managed environment reach the required company files and systems while preserving accountable access and a human approval point?

For readers outside China

• Availability: Qianwen Office officially began public beta testing on August 3. Individual and enterprise users can try it through its official website's web version and PC client. Availability outside China is not disclosed in sources.
• Pricing: WorkBuddy's personal standard plan costs 70 yuan per month with continuous monthly billing. Both WorkBuddy and Qianwen Office give new registrants 2,000 credits and provide 100 credits for daily logins. Qianwen Office's paid pricing is not disclosed in sources.
• Closest Western equivalents: The source material does not identify Western equivalents.
• Data residency: The source material does not cover data residency, data-location guarantees, cross-border transfer, encryption, retention, or compliance terms. Qianwen Office includes a cloud database, and Nano Work is described as providing a cloud office environment, but the sources do not say where data or workloads are hosted.

Sources

• woshipm 千问办公，阿里准备用来打WorkBuddy么？ https://woshipm.com/ai/6439860.html
• woshipm 千问办公硬刚WorkBuddy，5轮对比测评后，结果有点意外 https://woshipm.com/ai/6436882.html
• geekpark 业内首个！「千问办公」推出「多人工作台」 https://geekpark.net/news/369934
• 36kr 360的奥德赛时刻：又一家公司冲进AI办公 https://36kr.com/p/3972799003783680