Tencent Tang Daosheng: Harnessing engineering capabilities is a key factor for AI implementation

LightningPacketLoss · 2026-03-27T06:50:52+00:00

At the Tencent Cloud Shanghai Summit, Tang Daosheng pointed out that AI implementation is an engineering challenge, emphasizing the importance of systems engineering methods. Tencent Cloud enhances model deployment capabilities through the Intelligent Agent Development Platform ADP and the Claw system, providing knowledge bases and secure sandbox solutions, improving reinforcement learning efficiency, and achieving efficient learning and accumulation.

LightningPacketLoss

2026-03-27 06:50:52

Abstract generation in progress

On March 27, at the Tencent Cloud Shanghai Summit, Tencent Group Senior Executive Vice President and CEO of the Cloud and Smart Industries Group, Tang Daosheng, stated that the implementation of AI is not just an algorithmic problem, but also an engineering challenge. With the same model capabilities, different Harnesses, which include system engineering methods such as tool invocation, hierarchical context engineering, long memory management, and workflow design, will all affect the actual effectiveness of AI implementation.

Tencent Cloud enhances model implementation capabilities from multiple dimensions. First is the Tencent Cloud Intelligent Agent Development Platform (ADP), which connects agents to a specialized “library” through RAG and knowledge base capabilities, allowing industry experts to always be online and continuously accumulate the company’s operational know-how. Then there is Claw, running in the Agent Runtime’s secure sandbox. Claw serves as the neural hub of this intelligent system, continuously learning and accumulating the ability to connect with external systems by discovering and downloading Skills from the skill library, and using large models to send and receive commands externally, triggering actions.

This Agent Runtime sandbox solution can also be used for verifying the results of large model reinforcement learning programs, capable of launching over 100,000 container sandboxes in under one minute, with startup speeds in the milliseconds range, and being destroyed after use, significantly enhancing the training efficiency of reinforcement learning.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

1 Likes