WorkBuddy狂奔230天：重构办公，也被办公重构

Key information

1. WorkBuddy completed 9 billion user task requests in nine months since launch; the enterprise edition has been deployed in more than 50 industries and covers 50% of China's top 500 enterprises; average daily Token calls exceeded 500 trillion, growing more than 5,000x in two years.

2. Tencent Cloud launched Buddy One, Tencent Wangpan, and enterprise-grade Agent governance and runtime infrastructure, shifting office collaboration from file circulation to human-AI co-creation; Wangpan becomes the unified context and memory for Agents.

Practical takeaways

1. Enterprises choosing Agents must pass four gates: runs depends on MCP and connectors linking business systems; trustworthy depends on identity governance, sensitive-data masking, execution protection, and full-process auditing; measurable depends on an observability center to see where Tokens go; deployable depends on real-world scenario ROI.

2. Reference cases: Taikang uses AI for 24/7 insurance-agent practice and pre-underwriting with 95% accuracy; Yili raised low-temperature demand forecasting from 70% to 90% and improved store order fulfillment by 15%; Suning used trajectory analysis to cut Token consumption by 10%; Qiyuan Robotics connected more than 4,000 embodied skills.

3. The platform emphasizes openness without lock-in; enterprises can switch workbenches, models, and engines.

Brand marketing and user operations

1. Enterprise Agents are moving from individual productivity to organizational restructuring; Taikang distills the knowledge of high-performing trainers into AI, and 300,000 to 400,000 insurance agents practice with AI 24/7 via mobile phones, showing that brand training, service, and user outreach can be handled by Agents.

2. Yili uses AI models and agents to forecast demand for low-temperature products, lifting accuracy from 70% to 90% and improving terminal-store order fulfillment by 15%, showing that brands can use AI to improve channel fulfillment and freshness experience.

Product R&D and channel collaboration

1. Yili replaces dealers' individual safety stock with brand-shared inventory, showing that brands can use AI to redesign inventory and channel collaboration and reduce channel fragmentation.

2. After Qiyuan Robotics connected to WorkBuddy, it integrated more than 4,000 embodied skills across Tencent Meeting, WeRead, QQ Music, and other scenarios, showing that product R&D can extend AI capabilities to hardware and multiple scenarios.

Price competition and investment decisions

1. Enterprise Agents face high costs, data silos, and hard-to-calculate ROI; the observability center let H3C see team usage depth and reset Token incentives, while Suning used trajectory analysis to cut Token consumption by 10%, suggesting brands should account clearly for AI investment.

2. WorkBuddy supports third-party IM, third-party models, and enterprise self-developed Agents, emphasizing openness without lock-in; brands should evaluate compatibility and autonomy when choosing.

Policy and market opportunities

The article does not provide direct support policies, but it offers growth-market signals: WorkBuddy's enterprise edition has landed in more than 50 industries and covers 50% of China's top 500 enterprises; Tencent's overseas active customers doubled over the past year, covering more than 150 countries, and it plans to expand data centers in Asia, the Middle East, Europe, and the Americas.

1. Opportunity signals: enterprise Agents are entering an explosive phase, with average daily Token calls exceeding 500 trillion and growth of more than 5,000x in two years; sellers can watch demand in enterprise office, knowledge bases, business-system connections, and multi-model access.

2. Collaboration methods: WorkBuddy supports third-party IM, third-party models, and enterprise self-developed Agents, and opens through MCP and connectors; sellers can look for entry points around compatibility, connectors, and industry Agents.

Risks and lessons

1. Risk signals: enterprise Agents have problems such as high costs, data silos, permission security, and unclear ROI; deployment cannot be a blind gamble.

2. Lessons: use an observability center to see where Tokens are spent; H3C reset Token incentives based on this, and Suning used trajectory analysis to cut Token consumption by 10%; first pass the four gates of runs, trustworthy, measurable, and deployable before scaling.

3. Business model change: big tech is shifting from API calls and selling compute to organization-level productivity systems; WorkBuddy is positioned as the enterprise Agent super control console.

Production planning and product design

1. The Yili case offers direct lessons for factories: using AI models and agents to forecast demand for low-temperature products raised accuracy from 70% to 90% and improved terminal-store order fulfillment by 15%; replacing dealers' individual safety stock with brand-shared inventory shows that production planning can use AI and inventory collaboration to reduce blindness.

2. Taikang used nearly 50,000 cases and a 1.8 million-entry medical knowledge graph for pre-underwriting and achieved 95% accuracy, showing that turning business knowledge into AI-callable assets can feed back into product design and rule design.

Business opportunities and digitalization

1. Enterprise Agents have landed in more than 50 industries and cover 50% of China's top 500 enterprises; office suites, Tencent Wangpan, Agent runtime infrastructure, and embodied skills all present industry opportunities; Qiyuan Robotics connected more than 4,000 embodied skills, showing that intelligent hardware can use AI as an external brain.

2. Digitalization lessons: for Agents to run, they need to connect ERP, CRM, and other business systems; to be trustworthy, they need identity governance, data masking, execution protection, and auditing; to be measurable, they need an observability center; factories advancing digitalization should not look only at point productivity gains but calculate ROI for organizations and business flows.

3. Buddy One uses WorkBuddy as the workbench to connect Tencent Docs, IMA, and Tencent Design, shifting collaboration from file circulation to human-AI co-creation; this can inform product design and collaborative process restructuring.

Industry trends and new technologies

1. AI office is shifting from a battle of 100 models to a battle of 100 Agents, and enterprise Agents are entering an explosive phase; WorkBuddy completed 9 billion task requests in nine months since launch, and its enterprise edition has landed in more than 50 industries, covering 50% of China's top 500 enterprises.

2. New technologies include MCP and connectors for linking business systems; IOA endpoint security for identity governance and sensitive-data masking; execution protection and full-process auditing; an observability center as an Agent dashboard; Tencent Wangpan as a unified context space; TokenHub for multiple models such as Hunyuan, DeepSeek, and Kimi; and Agent Runtime as an operating environment.

Customer pain points and solutions

1. Pain points: individual productivity gains do not equal organizational productivity gains; enterprises fear that permissions are not controlled and results are not validated, affecting an entire business line; costs are high, data silos exist, it is unclear where Tokens are spent, and ROI is hard to calculate.

2. Solutions: use Buddy One to connect the workbench and underlying tools, use Tencent Wangpan to accumulate task assets, and use the four gates of runs, trustworthy, measurable, and deployable; H3C reset Token incentives with the observability center, and Suning used trajectory analysis to cut Token consumption by 10%.

3. Service providers can design services around connectors, security auditing, observability, industry Agents, and open compatibility capabilities.

Platform needs and problems

1. Enterprises need an organization-level productivity system that can run, is trustworthy, measurable, and deployable; the problems are high Agent costs, data silos, permission security, and difficult ROI calculation, and individual productivity gains rarely become organizational productivity gains automatically.

2. Platforms cannot just sell APIs and compute; they must provide enterprise Agent governance and runtime infrastructure.

Latest platform moves

1. Tencent Cloud launched Buddy One and Tencent Wangpan, and uses WorkBuddy as the workbench to connect Tencent Docs, IMA, and Tencent Design; the bottom layer is a heterogeneous compute cluster and the multi-model TokenHub, the middle layer is Agent Runtime, and the upper layer is WorkBuddy, Buddy One, Wangpan, and an open Agent ecosystem.

2. In operations management, IOA is used for identity governance, sensitive-data masking, execution protection, and full-process auditing; an observability center is used to see where Tokens go, supporting Token incentives and trajectory analysis; the Suning case cut Token consumption by 10%.

Partner recruitment and trend risk avoidance

1. Open through MCP and connectors, support third-party IM, third-party models, and enterprise self-developed Agents, emphasize openness without lock-in, use compatibility to win enterprise-grade entry points, and secure the position of enterprise intelligent hub.

2. Overseas active customers doubled and cover more than 150 countries; it plans to expand data centers in Asia, the Middle East, Europe, and the Americas; platform operators need to avoid closed lock-in and talking only about model benchmark scores, and shift toward measurable ROI and ecosystem compatibility.

New industry dynamics and business models

1. AI To B is entering deep waters, and competition is shifting from the battle of 100 models, API calls, and selling compute to organization-level productivity systems; the Agent explosive phase is underscored by average daily Token calls of 500 trillion and more than 5,000x growth in two years, and IDC expects global enterprise-side AI spending to reach $940 billion in 2026.

2. Tencent has formed a three-layer ecosystem: the bottom layer is heterogeneous compute and multi-model access via TokenHub; the middle layer is Agent Runtime; the upper layer is WorkBuddy, Buddy One, Tencent Wangpan, and an open Agent ecosystem; WorkBuddy is positioned as the enterprise Agent super control console.

New problems and governance implications

1. The core new problem is that individual productivity gains do not equal organizational productivity gains; enterprise Agents are constrained by four gates: runs, trustworthy, measurable, and deployable, and face high costs, data silos, permission security, and difficult ROI calculation.

2. Governance implications: identity governance, sensitive-data masking, execution protection, and full-process auditing make Agents trustworthy; an observability center makes Tokens visible, with H3C resetting Token incentives and Suning cutting consumption by 10%; openness without lock-in supports third-party IM, models, and self-developed Agents.

3. Cases provide research material: Taikang's AI practice and pre-underwriting at 95% accuracy; Yili's demand forecasting rising from 70 to 90 and order fulfillment up 15%; Qiyuan Robotics connecting more than 4,000 embodied skills; overseas customers doubling and covering more than 150 countries, allowing observation of AI To B globalization.