Qianwen Office launches "Enterprise Context," the industry's first enterprise-grade agent product, along with the AI hardware QwenNote A2.

AI agents are moving from chat windows into enterprise organizations, and Alibaba's Qianwen Office is further completing the product chain from "understanding the business" to "being able to execute".

On September 22, Alibaba's Qianwen Office launched a series of enterprise-level AI Agent products at the 2026 Hangzhou Yunqi Conference, introducing features such as enterprise context, digital employee, collaboration space, and security center, thus improving the product system of agents from business understanding to execution and collaboration. On the hardware side, Qianwen Office also launched the QwenNote A2 AI Agent personal assistant.

Among them, Enterprise Context helps Agents obtain task-related information by integrating enterprise data, further building a "business-savvy" digital employee.

At the conference, Chen Yusen, Vice President of Alibaba Group and CEO of Qianwen Office, stated that "Qianwen Office is the industry's first enterprise-level agent product." He pointed out that for an agent to truly be implemented in enterprises, it must not only understand the business but also collaborate within the enterprise and organization like a person, while ensuring that every task execution has boundaries and is traceable.

In terms of hardware, Chen Yusen revealed that the previous generation DingTalk A1 had achieved sales of "hundreds of thousands of units" in China, and set a sales target of "tens of millions of units" for the new QwenNote A2, demonstrating Qwen Office's further layout in the AI hardware market.

Enterprise Context: Enabling Agents to Truly Understand Business Data

One of the core products released by Qianwen Office this time is "Enterprise Context", which is positioned as a context data management product for enterprises.

The core logic is that enterprises often accumulate large amounts of data, but when an agent executes a specific task, it doesn't need all the information, but rather the content most relevant to the current task. Enterprise context, through a built-in dedicated model, compresses and structures enterprise data at a low token cost, and updates it in real time as business changes, allowing the agent to make judgments based on the latest business information.

This product directly addresses the issues of data silos and information overload in enterprise AI implementation. Based on enterprise context and Qianwen Office's agent hosting capabilities, enterprises can build "business-savvy" digital employees and deploy them on instant messaging platforms such as DingTalk to collaborate with employees to complete tasks.

Digital Employees and Collaborative Spaces: Agents Moving Towards Organizational Operations

At the digital employee level, Qianwen Office gives Agent a complete "organizational identity": it has a name, department, person in charge and job responsibilities, as well as authorization scope and lifecycle management . It can be identified by employees, authorized by the system, and also deactivated by administrators. All execution records can be traced back to the corresponding identity.

Meanwhile, Qianwen Office launched a "Collaboration" feature, allowing enterprises to create collaborative spaces that unify and link information scattered across group chats, documents, and knowledge bases, and add team members and digital employees as needed. Within this space, employees can directly discuss with colleagues or agents, and agents can also communicate and collaborate with each other. Task outputs and progress are uniformly stored in the space, reducing the repeated flow of information between different groups.

For work scenarios that are not suitable for completion within a chat window, Qianwen Office has also launched an "Application" function, supporting enterprises and SaaS vendors to integrate their business systems as applications, while retaining the original interface and data and adding Agent capabilities. Users can directly modify and save AI-generated results within their business systems without having to switch between multiple tools.

Security Center: Managing Agent Risks at the Architectural Level

As AI moves from "answering questions" to "performing operations," security boundaries have become a key consideration for enterprises adopting agents. Chen Yusen stated that enterprises should not only add defenses around the system's perimeter, but should also consider risks such as access violations, data leaks, and accidental operations from the initial architecture design stage.

The newly released Qianwen Office Security Center can uniformly manage agent data access, tool calls, and task execution. Specific mechanisms include: defining operational boundaries through a sandbox before execution; approving and blocking high-risk operations during execution; and handling anomalies through operation logging and recovery mechanisms.

Combined with DingTalk's enterprise security system, this product also supports the granting of least privileges, identification of sensitive information, and full-process auditing, enabling enterprises to trace the task initiator, the data used by the Agent, and the specific execution operations.

QwenNote A2: AI hardware with privacy protection as its differentiating selling point

On the hardware front, Qwen Office released the AI Agent personal assistant QwenNote A2, which emphasizes "true AI, no recording" to address privacy concerns raised by traditional AI recording products.

In terms of privacy protection, the QwenNote A2 adopts a "single transcription, self-destructing" design: the device does not record audio by default; after a conversation ends, the voice is transcribed into text in the cloud, and the original recording is permanently deleted; a small green light illuminates when the device is working, informing the other party that it is only taking notes and not recording audio. The device weighs 67 grams, is equipped with a six-microphone array, supports independent network connectivity and agent access, and can operate without relying on a mobile phone.

In terms of usage scenarios, QwenNote A2 supports converting offline communication into cloud tasks with a single click. For example, after a salesperson visits a client, they can use voice commands to synchronize the communication content to QwenNote Office, automatically generating a PPT or extracting key information.

Chen Yusen revealed that the previous generation product, DingTalk A1, had already achieved sales of "hundreds of thousands of units" in China. He believes that AI recording is just a small category, while AI notes, AI intelligent conferencing, and other products have a much larger application space and could very well achieve sales of "tens of millions of units" in the future.

Qwen Intelligence: The Thousand Questions Big Model Further Expands Personal Intelligence

At the same time, Qwen Intelligence, a full-stack AI mobile solution, was also released, providing mobile phone manufacturers with model, platform and scenario solutions to help mobile agents move from simple question answering to complex task execution.

Addressing the core needs of mobile agents, Qwen Intelligence has launched three solutions: Mobile Planner Agent, responsible for understanding requirements, breaking down tasks, and dynamically planning; Mobile-Use Agent, responsible for the actual operation of the mobile phone; and Mobile Creative Agent, geared towards image creation, transforming a user's single-sentence requirement into a complete creative process. Official data shows that the Mobile-Use Agent achieved a 90% end-to-end task success rate in real-world testing, reducing single task completion time from 83.6 seconds to 59.5 seconds.

Currently, Qwen Intelligence has begun to see practical applications. Alibaba, in collaboration with Honor, is using Qwen Intelligence and Honor Magic OS as a foundation to jointly create vertical domain models and solutions for next-generation AI smartphones. Official data shows that the solution achieves an overall task accuracy of 91.8%, can execute more than 100 steps for complex long-term tasks, and has an end-to-end service closed-loop rate of 90%.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.