On June 5, at the 2026 Tencent Cloud AI Industrial Application Conference, the Tencent Cloud Intelligent Agent Development Platform ADP officially launched its 4.0 version, upgrading to an enterprise-level AgentOps platform. The new version adds support for the Claw mode of the Agentic Loop, and through Connector, Skills, knowledge base, MCP, and Agent Portal, it connects the entire lifecycle of enterprise-level Agent construction, connection, distribution, and governance, helping enterprises achieve large-scale industrial deployment of Agents.

image.png

Wu Yunsheng, Vice President of Tencent Cloud

Upgrade Claw Mode, Lower the Bar for Enterprise Agent Construction

One of the core upgrades in ADP 4.0 is the addition of the Claw mode, which supports the Agentic Loop mechanism, on top of the existing three construction modes: LLM+RAG, workflow, and Multi-Agent.

The Claw mode is designed for more complex and longer business tasks. It allows creating Agents that can autonomously code and run in a cloud sandbox, call enterprise Skills, and perform long-term tasks. After creation, these Agents can be directly integrated into enterprise business systems via API interfaces and reach employees and customers through channels like WeCom and WeChat.

In terms of operation, creators do not need to configure forms; they just need to describe their needs in natural language, and the platform will automatically generate prompts, mount the knowledge base, configure tools, and orchestrate workflows. For standardized business scenarios, applications can also be quickly created based on templates, with Agent capabilities continuously adjusted during the conversation.

To ensure that Agents reliably complete complex tasks, ADP 4.0 also supports bidirectional calls between Agents and Workflows. Agents can call workflows to handle clearly defined, deterministic business processes, while workflows can trigger Agents when encountering unstructured judgments, complex analysis, or open-ended tasks. This mechanism enables enterprises to establish more flexible collaboration between deterministic processes and intelligent decision-making.

The Claw mode has been validated in many scenarios. For example, in the Tencent Industrial Quality Inspection Platform TI-AOI, traditional visual inspection heavily relies on engineers' on-site experience, requiring switching between multiple pages for data checking, log viewing, model judgment, and parameter adjustment. The quality inspection Agent built using the Claw mode can close the loop for processes such as "identifying defects, considering standards, performing inspections, checking logs, and adjusting models." Engineers only need to input "Help me check if this batch of data can be trained," and the Agent can automatically perform data health checks, assess training feasibility, and provide subsequent optimization suggestions.

Connect Enterprise Systems, Knowledge, and Tools, Embedding Agents in Workflows

For enterprise-level Agents to enter production environments, they must be integrated with existing enterprise systems, data, and processes. ADP 4.0 supports connecting to enterprise systems, data, and workflows through Connectors, Skills, knowledge bases, and the MCP standard protocol, transforming scattered business resources into reusable AI assets.

In terms of Connectors, ADP 4.0 initially launches nearly 40 selected Connectors, supporting integration with high-frequency office and business systems such as CRM, ERP, OA, tickets, customer service, project collaboration, enterprise cloud storage, knowledge base, and document systems. Enterprises no longer need to manually transfer data or reorganize materials; Agents can directly read, retrieve, and call information from business systems and further perform operations such as querying, analyzing, generating, and transferring data.

In terms of Skills, ADP 4.0 upgrades the Skills Square, now supporting over 150 Skills. Business teams or developers can also package custom Skills and submit them as shared plugins for the enterprise. After security checks and approval, these Skills will be available in the enterprise area for use by Agents and smart workstations.

At the same time, ADP 4.0 has accumulated over 50 scenario-based templates and industry-specific applications. For example, in the retail industry, the platform provides ready-to-use business templates such as marketing short video creation, copywriting assistants, and e-commerce Q&A review. Enterprises can either quickly build applications based on these templates or customize them by combining their own processes, knowledge bases, and tool capabilities, shortening the cycle from zero to one.

In terms of the model ecosystem, ADP 4.0 supports the unified access of mainstream large models and enterprise-built models, and natively complies with the MCP standard protocol, enabling one-time configuration and multi-use application.

Additionally, the ADP 4.0 Intelligent Agent Workbench supports access through various entry points, including the platform, browser, Office, IM channels, API/SDK, allowing Agents to be embedded into employees’ daily work scenarios.

Full-Chain Governance from Development to Launch, Passing the “Security Check” Before Going Live

When intelligent agents enter the production system, security and governance become essential requirements. ADP 4.0 pushes governance capabilities to the source of development. By focusing on permission management, Skills governance, operational observation, and deployment compliance, it establishes a comprehensive security control system covering the entire lifecycle of Agents.

In terms of permission management, ADP 4.0 supports a hierarchical permission architecture at the enterprise, space, and application levels, combined with RBAC role permission matrices, achieving dual-dimensional isolation of functional and data permissions. Enterprises can configure access ranges based on organizational structure, departments, positions, and roles, ensuring clear permission boundaries between different teams, applications, and knowledge bases.

In terms of Skills governance, the platform supports a closed-loop lifecycle from submission, security checks, approval, listing, distribution, to usage monitoring. Self-created Skills submitted by employees must pass code static scanning, data access, network outbound, and dependency whitelist security checks, along with multi-level approvals before being used in the enterprise area. Through governance upgrades, Skills are no longer personal tools but enterprise assets that have been approved, shared, and scheduled.

In terms of Agent observation and governance, ADP provides full-chain observation capabilities. Enterprises can centrally manage Agents across different platforms and business scenarios through the Agent Portal; by using business dashboards and resource dashboards, they can view key metrics such as call volume, activity level, response quality, operational cost, and error reports in real-time, quickly locate causes, and optimize performance.

To meet the compliance requirements of different industries, ADP 4.0 supports four deployment modes: public cloud, private cloud, hybrid cloud, and dedicated cloud. At the same time, the private deployment solution for the intelligent workbench and security sandbox will also be released with version iterations, supporting secure execution of code, calling Skills, and executing long-term tasks within the customer's internal network.

At the event, Tencent Cloud upgraded its ecological strategy for ADP, collaborating with ISV partners to co-build vertical industry Agents, promoting large-scale delivery of industry-specific intelligent agents. At the same time, it opened up standard products, development frameworks, and toolchains for resellers and delivery partners, helping enterprises quickly achieve AI efficiency enhancement and personalized implementation. For example, with Zhaogangwang, both parties have jointly developed a steel trading industry Agent covering the entire supply chain, including inquiry, procurement, warehousing, and logistics, and extended the model to fields such as bulk goods, manufacturing, and energy.

Wu Yunsheng, Vice President of Tencent Cloud and Head of the Tencent Cloud Intelligent Agent Development Platform, stated that enterprise-level Agents are not about who builds them faster, but about who can operate them stably, securely, and continuously in the business environment. ADP aims to provide an enterprise-grade AgentOps foundation, enabling enterprises to quickly build Agents, integrate them into systems, distribute them to the front line, and continuously operate them under permission, security, evaluation, and observation systems, thereby unleashing productivity value.