Leike Technology AI Hardware Group | Editor: Tianxing | Supervisor: Luo Chao
AI Agents have become the core driver for "unlocking productivity" across industries in 2026. From IT giants to individual bloggers, everyone is "unlocking" richer identities and capabilities driven by AI Agents. At the "2026 Tencent Global Digital Ecosystem Conference" held on October 29, Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud & Smart Industries Business Group (CSIG), stated that since its launch in March this year, WorkBuddy has cumulatively completed 9 billion task requests. The enterprise version has been deployed in over 50 industries, covering 50% of China's Top 500 Enterprises and 60% of China's Top 100 Enterprises.
However, the issue lies in the fact that employees using AI does not mean enterprises are truly utilizing AI. This statement may seem convoluted: Take Leike Technology as an example; we are also heavy users of WorkBuddy. Specific tasks such as article statistics, meeting transcriptions, and travel itinerary arrangements all leverage AI. But from a hierarchical perspective, these AI work scenarios largely treat WorkBuddy-type AI Agents as personal assistants, rather than having Agents directly serve the enterprise's workflow in ways that align with AI strengths.
From an enterprise management perspective, in this scenario, AI Agents primarily reduce the burden on employees rather than reducing the burden on the enterprise or optimizing production structures. In other words, under this working model, AI Agents only demonstrate their value to employees without providing value to the enterprise itself.
For this reason, Leike Technology believes that the true difficulty in developing ToB businesses for AI Agents lies in how to integrate AI Agents into the most core workflows of enterprises, creating "native" Agent workflows. This allows Agents to help enterprises find new growth points beyond cost reduction and efficiency improvement, which is also a key direction of Tencent's AI ecosystem upgrade at the 2026 Tencent Global Digital Ecosystem Conference.
However, upgrades in technology and products alone are insufficient. What bottlenecks and difficulties remain when integrating Agents into enterprise workflows? At the 2026 Tencent Global Digital Ecosystem Conference, Leike Technology spoke with Chen Liang, General Manager of Internet Business Technology at Tencent Cloud; Ding Junwei, Head of the "Titanium Pole" Large Model at Tec-Do; and Li Tianyu, Co-founder and CEO of Source Future. They discussed the practical deployment of AI Agents, attempting to identify the difficulties and solutions for enterprises to "use AI effectively" from the perspectives of AI service providers and users.
To Deploy in Enterprises, Agents Must Deliver "Returns"
To create so-called native Agent workflows, Agent services must first prove their value to enterprises. During the discussion, Chen Liang, General Manager of Internet Business Technology at Tencent Cloud, noted that "Enterprises are willing to invest in AI because they fundamentally want to see returns."
After all, investment in software processes is not like traditional hardware or software investments where effects can be seen immediately. From the service provider's perspective, to encourage enterprises to invest in AI, Agents must deliver sufficiently high returns. These returns can come from costs saved through Agent-driven production models or from increased business revenue.
When asked about the biggest challenge in promoting the scaled application of agent platforms in enterprise businesses, Chen Liang acknowledged that the hardest part of promotion is cost. It is important to note that this cost includes not only the expense of restructuring production lines for AI but also subsequent usage costs.
During his sharing, Chen Liang mentioned a very interesting phenomenon: In the past, internet business innovations could use economies of scale to amortize marginal costs, but this method is difficult to apply to the usage of AI models.
He pointed out, "Until now, regarding model usage, costs for new business innovations including enterprise applications tend to be linear—because more users and more access times result in more Tokens." Token usage directly impacts operational costs for enterprises.
In response, Tencent is actively seeking ways to reduce enterprise AI costs, such as using domestic open-source models and compressing context windows.
From Leike Technology's perspective, this promotion strategy, completely different from the ToC market, explains from the supplier's angle why Tencent Cloud is building its own full-stack AI engine while accelerating the adaptation of Agent platforms like WorkBuddy to various industries. Unlike the ToC market, in the ToB industry, every sector is distinct, and the way AI Agents provide value differs across industries and even within individual enterprises.
As a service provider, Tencent Cloud cannot serve all industries with a single AI component. Therefore, in Tencent Cloud's Agent solution landscape, we see sufficient AI service components provided for different needs across various industries. This allows enterprises to build their own Tencent Cloud Agent deployment plans based on actual requirements, enabling AI to adapt to the characteristics of different industries and enterprises, thereby building "native AI" production lines at lower costs.
Providing AI Infrastructure, Tencent Helps Enterprises Spend Money Where It Counts
Take the cooperation between Tencent Cloud and Source Future as an example. The problems faced by embodied intelligence companies are entirely different from those of internet companies. To introduce briefly, Source Future focuses on base models for general-purpose whole-body humanoid robots, enabling robots to coordinate limbs and torso to complete complex whole-body tasks.
As a "user" of Tencent Cloud in the ToB domain, Li Tianyu, Co-founder and CEO of Source Future, mentioned, "Our current training of 'cerebrum' and 'cerebellum', along with the entire data storage and processing pipeline, and the link from data to real-machine deployment, largely runs on Tencent Cloud."
The biggest difference between embodied intelligence and large language models lies in the data layer. According to Li Tianyu, "There are no ready-made physical trajectories available on the internet. Models must complete data collection and post-training independently within their respective scenarios. One scenario requires one set of data; switching tasks means starting over."
Besides data issues, the computing power cost problem mentioned earlier cannot be ignored. Training the cerebrum and cerebellum consumes significant computing power, and models must undergo simulation verification before entering real environments. For a robotics company, building its own server room to handle these two aspects is not cost-effective, as money should be spent on models and the robot body itself. Consequently, Source Future moved the entire chain to Tencent Cloud.
From the perspective of Tencent Cloud as a service provider, the data collection, cleaning, labeling required by the embodied intelligence industry, the cloud training clusters needed for cerebrum and cerebellum training, and the simulation and validation of models in physical environments represent areas of cooperation between Tencent Cloud and the embodied intelligence industry.
For instance, during the conference, Tencent officially launched Tairos, the first modular software platform for embodied intelligence in China, providing enterprises with cloud services, large models, training environments, and a full SDK suite.
According to Li Qiang, Vice President of Tencent Group and President of Enterprise Business, the platform currently serves 90% of leading embodied intelligence enterprises; additionally, the WorkBuddy open platform supports the integration of various robot hardware. In the field of embodied intelligence, Tencent Cloud is well-prepared.
AI Transforms Marketing: Mass-Produced Content Cannot Withstand Budget Scrutiny
Having discussed the relatively emerging industry of embodied intelligence, let us examine how AI capabilities represented by Tencent Cloud Agents are transforming the traditional industry of "marketing," which has existed for over 100 years.
According to Ding Junwei, Head of the "Titanium Pole" Large Model at Tec-Do, at the current stage, Agents or large models can address issues of content production efficiency and execution efficiency in the marketing industry, replacing extensive manual labor through batch content production, creative insights, and delivery optimization.
However, these improvements in efficiency have brought a new problem: "Ad placement ultimately depends on budget decisions and actual spending. Even if a thousand materials are produced, perhaps only a hundred will be used. Therefore, human attention has become a bottleneck in the advertising and marketing field."
To solve this problem, Tec-Do launched the Navos intelligent agent and performed post-training on open-source models specifically for the advertising and marketing domain. However, during the exchange, Ding Junwei admitted, "(Currently) the capabilities of the models have not yet reached the level where they can truly replace humans in making decisions."
From Leike Technology's perspective, this phenomenon addresses concerns about AI replacing human work. Thanks to rapidly evolving large model technologies, many advertising companies have begun using AI to mass-produce marketing content, flooding the space with visual formulas to capture user attention and establish brand awareness for clients. For example, the simulation management game "Crazy Water World," which went viral online in recent months, successfully broke out of niche circles due to its imaginative yet misleading AI promotional videos.
But we know that consumer time is limited, and enterprise budgets are also limited: the former determines that poorly made content can only gain short-term attention by appealing to aesthetic ugliness, failing to provide sustained attraction for products; the latter implies that enterprises will inevitably allocate limited budgets to the best content, forcing content producers to improve quality and freshness.
To summarize in Ding Junwei's words: "Poorly made content definitely cannot withstand the test of budgets."
Delegate Execution to AI, Keep Thinking for Yourself
During the exchange, regarding the division of labor between AI and humans within enterprises, Chen Liang added a concise viewpoint: "We can delegate execution and work to AI, but we should not delegate thinking to AI." In his view, anxiety is inevitable when humans interact with AI; to overcome this anxiety, we must keep the process of thinking for ourselves.
Leike Technology believes that Chen Liang's summary of the division of labor between humans and AI also addresses the initial topic of this exchange from another angle: Returning to the question posed at the beginning of this article—how can enterprises integrate AI Agents into their most core workflows to find new growth points beyond "simple" cost reduction and efficiency improvement?
First, we must clarify that for enterprises, humans still possess irreplaceable value; the way AI Agents enter enterprises should not be to replace employees with higher efficiency, but to serve employees with higher efficiency, achieving human-AI co-creation.
Under this premise, what enterprises should do to embrace AI is not to replace humans with AI within existing work systems, but to return to the core values of the enterprise. First, understand what services the enterprise provides to customers and what value it creates; then consider how to leverage the potential of AI Agent platforms to allow employees to demonstrate higher value in areas such as creativity, emotion, thinking, and decision-making, where AI cannot tread.
Taking Tencent's recent AI upgrade as an example, the integrated suite Buddy One directly connects AI tools such as WorkBuddy, Tencent Docs, Tencent Meeting, Tencent Drive, and ima, serving different AI scenarios with a unified WorkBuddy Agent foundation. Leike Technology believes that this approach allows Buddy One to adapt to the actual workflows of different industries and enterprises, turning AI tools into the "lubricant" for every step of enterprise production and operations, thereby improving overall enterprise efficiency.
More remarkably, this strategy was not devised in isolation in a conference room but was explored by internal departments of Tencent and external partners in actual work. Backed by the massive Tencent ecosystem, Tencent Cloud can adjust and optimize its AI Agent capability components based on frontline experience in industries such as finance, education, manufacturing, and retail, better serving users.
Wang Huixing, Vice President of Tencent Group and CTO of Tencent Cloud, mentioned at the conference, "Tencent WorkBuddy Enterprise Edition has been deployed in over 50 industries including government affairs, education, retail, finance, media, and mobility, covering 50% of China's Top 500 Enterprises." It can be said that this AI Agent strategy, derived from the frontline and serving the frontline, is the greatest difference between Tencent Cloud and other AI service providers.
At this moment in 2026, characterized by an "explosion" of AI capabilities, Tencent's ultimate goal is to provide each enterprise with open platform capabilities, enabling them to go further in the AI era. At the 2026 Tencent Global Digital Ecosystem Conference, Tencent has handed its AI toolbox to enterprises; whether they can use these AI tools well to grow strong depends on the enterprises themselves.