As the key battleground in AI competition shifts from "model capabilities" to "engineering application capabilities," AI agents have taken center stage at this year's World Artificial Intelligence Conference (WAIC), supplanting large language models as the main focus. At WAIC, TENCENT showcased its comprehensive portfolio of AI agents and launched the overseas version of its upgraded enterprise-grade intelligent agent development platform, ADP 4.0.
A key question arises: in a context where personal AI agents like WorkBuddy are already widely adopted, why should enterprises go through the effort of building their own dedicated agents? In a discussion with media including Wall Street Insights, Tencent Cloud Vice President Wu Yunsheng not only explained the fundamental reasons for the existence of specialized agents but also highlighted several hurdles enterprises still face when implementing them.
Wu Yunsheng's response centered on one word: control. He contrasted two usage scenarios. When an employee uses WorkBuddy locally to create an effective sales process or integrate certain systems, enterprise managers might have no visibility into it, preventing colleagues from reusing it. However, if an employee creates a "sales assistant" on the ADP platform, it can be published with one click, allowing 30 employees to directly utilize it. Management gains clear visibility into usage and permissions, and the agent can even be embedded into the company's existing CRM or OA systems.
In essence, while personal and team productivity tools are well-served by convenient desktop agents, deploying agents within core business and operational contexts imposes much higher requirements on their construction, with a greater emphasis on integration and manageability. In such cases, a "cloud-edge integrated" approach based on ADP is more suitable.
Naturally, enterprises also encounter challenges when building their own agents. Wu Yunsheng noted that a significant current difficulty in agent deployment is interfacing with legacy systems. Vast amounts of data and processes are locked within these older systems, and many companies that previously underinvested in digitalization lack mature APIs or MCP interfaces for their legacy infrastructure to be called upon.
"Legacy systems will also undergo changes driven by current trends, as they were originally designed for humans—their interfaces, interactions, and outputs were meant for people to see and use," Wu Yunsheng stated. "The design philosophy for future systems must consider being built for Agents." He explained that the system design principles for Agents differ fundamentally from those for human users.
With advancements in Agent Loop capabilities, tasks that were previously limited by retrieval or workflow constraints are now becoming viable for implementation. A closely watched question is: will Agents cannibalize existing SaaS offerings? Wu Yunsheng's answer is "no." In his view, Agents act more as an additional layer of capability on top of existing systems, designed to connect and orchestrate, not to rebuild from scratch. The real value addition is enabling legacy systems, which have accumulated decades of data and processes, to develop interfaces that can be invoked by Agents.
Targeting the demand for enterprise agent development, TENCENT upgraded its large model knowledge engine to the Tencent Cloud Intelligent Agent Development Platform (ADP) in May 2025. Wu Yunsheng revealed that after over a year of development and with the emergence of new technologies this year, client acceptance of Agents has significantly increased. The demand for Tencent Cloud ADP has also seen substantial growth, at least doubling year-on-year compared to last year.
It is reported that Tencent Cloud ADP has now been deployed across more than 30 industries, covering business scenarios such as intelligent customer service, knowledge management, and media production. Focusing on the principles of being "deployable, scalable, and operable," Tencent Cloud ADP continues to strengthen its foundational capabilities. The latest ADP 4.0 upgrade is positioned as an AgentOps platform, providing enterprises with a one-stop service for building, distributing, and governing cloud-based intelligent agents.
ADP 4.0 has achieved bidirectional calling between Agents and Workflows, balancing open intent with rigorous processes. Furthermore, through OpenAPI, ADP exposes capabilities from modules such as spaces, applications, skills, plugins, knowledge bases, documents, and sessions, making it easier for enterprises to embed agent capabilities into their existing business processes.
Wu Yunsheng also emphasized that the competition for enterprise-grade agents is not about who builds them fastest, but about which agents can run more stably and securely in live business environments, forming an operational system that continuously creates value. Both capability and knowledge are indispensable. He believes the application of TENCENT's LeXiang Agentic knowledge base is also crucial. "While ADP addresses how to build, run, and manage agents, the knowledge base solves the questions of what the agent knows, what it believes, and what it uses."
Comments