千问办公束骏亮：AI办公的终点、赛点与胜负手

Bottom line: AI-powered office work is rapidly reshaping how we work, but adoption is difficult. Both individuals and enterprises can find new opportunities in context engineering and new hardware products.

1. Market opportunity: The AI office market is projected to grow to RMB 44.9 billion by 2026 and exceed RMB 332 billion by 2029, showing explosive momentum.

2. Real-world friction: 87% of executives report significant obstacles in scaling agentic AI, and 74% of organizations have not expanded AI value beyond pilots—most people and companies are still figuring out how to use AI effectively.

3. Critical path: No matter how powerful the model, it needs usable context. The average enterprise holds about 150 PB of internal data, while model context windows are around 1 million tokens. The core challenge is compressing and connecting data into knowledge AI can call upon. Qwen Office has cut the cost of such solutions from RMB 1 million and 10 days to RMB 5,000 and 3 days.

4. Hardware entry point: The QwenNote A2 can record conversations and automatically generate to-dos without requiring users to open a phone or computer. Hardware is becoming a gateway for capturing context and executing tasks.

Bottom line: The AI office race is shifting from model capability to context infrastructure. Brand owners' internal efficiency, product development, and B2B customer expansion strategies may all be reshaped as a result.

1. Product R&D: The essence of enterprise context is connection and compression—turning documents, meetings, group chats, and tacit knowledge into AI-friendly structured data. When developing internal systems or enterprise services, brands should prioritize whether data can be directly invoked by agents.

2. Cost and pricing: Qwen Office can build enterprise context in 3 days for RMB 5,000, a stark contrast to alternative proposals of 10 days and RMB 1 million. This shows that using extreme cost-performance to lower the barrier for customer trials is an effective competitive strategy.

3. User behavior: 87% of executives face significant obstacles in promoting agentic AI, and 74% of organizations have not extended AI value beyond pilots. This reflects that the real procurement pain point is not model parameters but whether AI can be embedded into actual business workflows.

4. Consumer trends: AI hardware is becoming an entry point connecting digital and physical worlds. For example, the QwenNote A2 can transcribe conversations and generate to-dos, potentially opening new office consumption habits through edge devices.

Bottom line: The AI office market is growing rapidly and the cost of building enterprise context is plummeting, creating incremental opportunities for service-oriented sellers—but there are also adoption barriers to watch.

1. Growth market: CIDC data shows the AI office market rising from RMB 21.2 billion in 2025 to RMB 44.9 billion in 2026, and exceeding RMB 332 billion by 2029—a clearly defined incremental track.

2. Changing demand: Mid-to-large enterprises hold an average of 150 PB of data, far exceeding a model's 1-million-token context window. What enterprises need is not a smarter large model but a service that turns massive data into usable agent context.

3. Business model: Qwen Office has turned enterprise context construction into a standardized deliverable: completed in 3 days for RMB 5,000. This proves low-cost replication is feasible. Sellers can enter through integration, implementation, training, and related services.

4. Risk and opportunity: 87% of executives report obstacles in promoting agentic AI, and 74% of organizations have failed to scale beyond pilots. Customers will become increasingly cautious. As model capabilities converge, the competitive focus shifts to context infrastructure; improving data compression and connection capabilities is key.

5. Collaboration potential: Qwen Office is not building a closed system. In the future, multiple AI products will share the same context infrastructure, offering third parties an entry point to build businesses on an open middle layer.

Bottom line: The AI office boom is driving not only software demand but also hardware entry points and on-device model deployment, creating opportunities for factories in product manufacturing and digital upgrades.

1. Hardware opportunity: Qwen Office released the QwenNote A2, defining hardware as a bridge between the digital and physical worlds. It can record conversations and automatically generate to-dos. Such AI hardware is suitable as a direction for finished devices or component businesses.

2. On-device model trend: The Qwen3.8-Flash model used in Qwen Office's daily operations can already run on hardware with 128 GB unified memory. This shows that on-device AI imposes new hardware requirements, potentially driving a new round of hardware design demand.

3. Digitalization insight: The essence of enterprise context is structuring documents, meetings, group chats, and tacit knowledge. Factory internal knowledge management also needs this kind of compression and connection, and AI can help reduce the cost of building the data foundation.

4. Business opportunity: The cost of building enterprise context has dropped from 10 days and RMB 1 million to 3 days and RMB 5,000, bringing AI office tools within reach for SMBs. If factories can provide devices for on-site data collection and physical-world information entry, they can complement the AI office ecosystem.

Bottom line: The AI office market is vast but difficult to implement. The opportunity for service providers lies in solving the connection between client data context and business scenarios.

1. Industry trend: The AI office market is projected to reach RMB 44.9 billion in 2026 and exceed RMB 332 billion by 2029. However, 87% of executives face significant obstacles in promoting agentic AI, and 74% of organizations have not extended AI value beyond pilots—a clear gap between demand and deployment.

2. Client pain point: Mid-to-large enterprises hold an average of 150 PB of internal data, while model context windows are only about 1 million tokens. Simply feeding all data in would blow the context limit or waste resources. Data sources are numerous and contradictory, so the coverage of abstraction algorithms becomes decisive.

3. Solution: Qwen Office connects documents, meetings, group chats, and tacit knowledge, and uses a graph structure to let AI explore context gradually. Delivery is achieved in 3 days for RMB 5,000, far below comparable proposals of 10 days and RMB 1 million.

4. Technical essentials: Multimodal models and self-developed chips make it possible to process complex data such as massive video. Construction must cover multiple software platforms, including DingTalk and Feishu, to build general capability.

5. Ecosystem opportunity: In the future, multiple AI products will share the same context infrastructure. Service providers can focus on integration, migration, and operations around the middle layer.

Bottom line: AI office is reshaping the enterprise software landscape. Platform players need to recognize the new trend where traditional software retreats to the base layer and shared context infrastructure becomes the battleground.

1. Platform positioning: According to Junliang Shu, traditional office software—including DingTalk, Salesforce, and SAP—will gradually retreat to become data and context foundations. Platforms should consider shifting from application-interaction models to data and context infrastructure models.

2. Open ecosystem: Qwen Office is not building a closed system. Enterprise context belongs to the enterprise and is decoupled from agent products. In the future, multiple AI products can share the same context infrastructure. Platforms can access this layer through open interfaces, enabling an ecosystem with multiple coexisting products.

3. Operations management: Not all processes are suitable for AI, given hallucination risks and token consumption. Platforms should keep stable, low-cost legacy processes for standardized tasks and let AI handle open-ended interaction tasks.

4. Risk mitigation: Clients themselves are not fully clear on their requirements, and converging model capabilities are reducing the advantage of pure brand volume. Platforms should focus on context connection and compression efficiency, with customer choice as the center.

Bottom line: AI office is not just a rapidly expanding industrial market; it is also reshaping software division of labor and enterprise data asset forms. New trends, new problems, and new business models deserve close observation.

1. New industrial trend: Traditional software (DingTalk, Salesforce, SAP) is gradually retreating to become the data and context foundation, while AI software takes over natural language interaction and task execution. The future enterprise AI ecosystem will feature multiple products sharing the same context infrastructure.

2. Core problem: The contradiction between massive enterprise data (an average of 150 PB) and model context windows (about 1 million tokens) makes context compression and connection central. However, data sources are numerous and contradictory, and the coverage of abstraction algorithms is not yet standardized.

3. Business model innovation: Large models have changed the ontology cost of building enterprise context. A high-cost business like Palantir's can now be completed in 3 days for RMB 5,000, shifting customized projects into standardized, replicable products—a typical sample of AI infrastructure-ization.

4. Empirical findings and implications: Projections of market size moving from RMB 21.2 billion to RMB 44.9 billion and then RMB 332 billion, along with 87% of executives facing obstacles and 74% of organizations failing to expand beyond pilots, indicate supply-demand mismatch and missing standards. Going forward, attention should be paid to context data ownership, cross-enterprise sharing mechanisms, and AI safety boundaries, which may give rise to new industry standards.