Tencent Cloud Upgrades Intelligent Agent Development Platform ADP and Releases AgentPortal 2.0

At China Resources Trust, the starting point of a bond trade may be nothing more than a message in a chat group.

Traders need to receive instructions across thousands of trading groups, identify the trade elements, and then enter them into the trading system. When clients ask about positions or trade progress, they also have to open the system to check and send the results back to the group.

This work is both tedious and unforgiving of errors.

Now, China Resources Trust has begun letting an agent take over part of this work. It enters the trading groups directly, identifies and breaks down trade instructions, enters them into the system, and can also check positions and trade progress and automatically reply to inquiries in the groups.

▲China Resources Trust Sunan

According to Tencent Cloud, this bond assistant, built on the agent development platform ADP, has raised its order-splitting accuracy from above 80% initially to 98%.

But at key points, China Resources Trust still retains manual confirmation.

Behind this is actually a non-consensus view in the industry: large models keep demonstrating how many complex tasks they can complete, but a model being able to correctly understand an instruction and an enterprise being willing to let it execute that instruction are two different things.

Reality may not be so simple. In August 2026, McKinsey released a global AI survey: 80% of respondents said AI improved their personal work efficiency, but only 37% believed AI had made a positive contribution to their company's profits. In September, McKinsey further noted that enterprises spend a large amount of time on task handoffs, coordination and waiting between different departments and systems.

This is also the challenge facing many small and medium-sized B2B AI companies focused on models or single-point applications. Model capabilities and product features can be iterated quickly, but entering core enterprise systems, gaining customer trust, and building a delivery system covering different industries require longer-term investment and accumulation.

On October 30, at the Tencent Global Digital Ecosystem Summit, Tencent Cloud upgraded the agent development platform ADP, released AgentPortal 2.0, and announced two-way integration with WorkBuddy.

With this upgrade, Tencent Cloud also further clarified ADP's positioning as a Service Agent, focusing on enterprise data and business scenarios, and driving agents deep into business processes to create real business value.

From enterprise business system integration to unified agent management, and then to industry scenarios and the delivery ecosystem, Tencent Cloud is trying to stake out the strategic high ground for the large-scale deployment of enterprise agents in advance.

1. Will AI dare to enter core business? Enterprise agents must first pass the management hurdle

From understanding instructions to executing trades, the first thing enterprise agents must solve is the issue of permissions and responsibility.

The reason China Resources Trust's bond assistant still retains manual confirmation and operation records is not hard to understand. A financial agent being able to query client positions does not mean it can view all clients' data; an agent responsible for handling business being able to call system interfaces does not mean it has the right to complete all operations.

For enterprises managing agents, the first thing to do is to draw clear lines around business permissions.

Responding to this need, Tencent Cloud ADP this time upgraded capabilities such as secure private network connections and tenant and permission isolation, supporting different agents in accessing internal enterprise resources such as databases and ERP according to permissions, and providing enterprise system connection capabilities uniformly to different modules such as sandboxes, workflows, MCP and connectors.

The same is true at Yunnan Baiyao: Yunnan Baiyao employees need to query internal policies, drug information and medical materials in their daily work, but the users and permission requirements for this knowledge differ.

To this end, Yunnan Baiyao, relying on ADP and Lexiang, refined permission management down to the employee ID. Employees ask questions through a single entry point, "Baiyao Ding," and the agent automatically determines the question type, routes it to the corresponding knowledge base, and then returns content according to the employee's actual permissions.

▲Yunnan Baiyao Dang Yabin

For employees, it is simply asking a question and getting an answer; but behind the scenes, multiple knowledge bases and a set of permission rules work together.

In addition to clarifying permissions, enterprises also cannot simply pursue full automation, as different businesses have different requirements for the degree of agent autonomy.

For example, businesses with fixed rules such as standard approvals and ticket handling often require a clear execution order and predictable results; work such as cross-department analysis and business decision support may require multiple agents to divide labor and collaborate.

To this end, Tencent Cloud ADP provides four modes: Claw, standard, multi-agent and workflow, enabling enterprises to choose different ways of building and executing agents according to actual business needs.

China Resources Trust adopted just such a combination: using Claw to handle autonomous planning tasks, using Workflow to execute business processes with clear rules, and always having traders confirm key trading steps.

This also echoes the industry's judgment on agent governance.

In May 2026, Gartner, a globally authoritative research institution, warned that if enterprises adopt a uniform governance approach for all agents, it could lead to two outcomes: simple tasks being overly restricted, while high-risk tasks receive excessive permissions.

And when agents truly enter the production environment, management cannot stop at launch.

Traditional software usually runs according to predetermined program rules, but model outputs do not have the same degree of determinism. The successful execution of one task does not mean the agent can stably complete similar tasks over the long term.

In particular, after enterprises begin to let agents run continuously, how to track each operation, monitor costs, locate faults, and correct problems in a timely manner once they arise all become new management needs.

Tencent Cloud ADP this time strengthened capabilities such as run records, cost and status monitoring, fault localization, and version rollback. At the same time, ADP can also use AI to analyze real business feedback and propose agent optimization suggestions, which then take effect after human review, allowing the Agent to be continuously evaluated and adjusted during operation.

For enterprise management Agents, they must be able to continuously execute tasks under clear business rules. Enterprise-grade engineering and governance capabilities are becoming the first layer of technical barriers beyond models. This upgrade of Tencent Cloud ADP is precisely centered on these management needs in production environments.

An enterprise is not simply the sum of a group of employees. Each person saving one hour does not mean the company can spend that same amount of time less to complete a business transaction.

But that is not how enterprise business operations work. A task often needs to pass through multiple stages, including the responsible person, regional teams, and frontline staff, flowing across different departments, personnel, and business systems.

In September 2026, McKinsey analyzed the production process of a Fortune 500 industrial enterprise in a study and found that most of the time spent in the entire business process was waiting for tasks to enter the next stage.

McKinsey called this the "Coordination Interface Tax."

For large enterprises, optimizing a single step in isolation may not solve the collaboration problems between different teams and systems. This also explains why employees use AI more and more frequently, but the overall efficiency and business returns of the enterprise may not grow in sync.

The collaborative exploration between Tencent Cloud and Yili provides another case. Yili aggregates regional sales, shopping guide performance, and market data, uses agents to assist in analyzing business conditions and forming strategies, and then extends the relevant capabilities to frontline shopping guide scenarios.

In this process, WorkBuddy undertakes part of the work for business personnel to use AI and conduct business analysis; the shopping guide agent built by ADP further enters consumer-facing sales scenarios, generating product recommendations and communication scripts based on customer needs.

According to Tencent Cloud, this combination increased the efficiency of the shopping guide operations team by more than 50%. At the same time, the shopping guide agent brought a 26% increase in sales orders in A/B testing. The former reflects internal efficiency, while the latter directly touches business results.

This is also the core difference between Personal Agent and Service Agent. Wu Yongjian, Vice President of Tencent Cloud, told us that Personal Agent starts more from personal use, while Service Agent needs to stand from the enterprise perspective and consider the permissions and security of different roles, as well as who the built capabilities can be shared with and how they are managed.

▲Wu Yongjian, Vice President of Tencent Cloud

Tencent Cloud's two-way connection between ADP and WorkBuddy this time is precisely an attempt to shorten the distance between personal AI use and enterprise business execution.

On the one hand, Agents and Skills built by enterprises through ADP can enter WorkBuddy for employees to call directly in daily office work; on the other hand, personal workflows explored and verified by employees through WorkBuddy can also be transformed into ADP agents and enter the enterprise's unified management and distribution system.

Personal Agents solve the problem of how employees use AI, while Service Agents further bring AI into enterprise business processes. The place where enterprise AI is truly likely to generate greater returns may precisely lie in reorganizing the relationships between these steps.

III. Agents are becoming easier to build, but enterprises are also worrying about "raising Agents"

The lowered threshold for agent development may cause enterprises to encounter the old problem of the software era again: there are too many Agents.

Different departments have their own knowledge bases, processes, and business needs, and may also build their own agents. Each Agent has value when viewed individually, but as the number continues to increase, enterprises may not know which Agents are actually running, who is responsible for maintenance, which functions are duplicated, and how many permissions each of them has.

In April 2026, Gartner called this phenomenon Agent Sprawl. A survey by the organization at that time showed that only 13% of surveyed organizations believed they had appropriate Agent governance mechanisms.

In response, Tencent Cloud adopted a "two-pronged" approach: ADP lowers the threshold for Agent development; AgentPortal 2.0 solves the problem of unified use and management of agents.

Among them, this ADP upgrade has two important changes: first, from conversational agent building to AI-native throughout the entire development process; second, from building agents to building and deploying complete business applications. ADP is upgrading from the 4.0 version's "agent building expert" to an "agent full-lifecycle expert."

In the past, business personnel might need to submit requirements to the technical team, and then developers would separately complete agent configuration, system integration, and application interface building. Now, part of the building work can be completed in the same workspace, and configuration, preview, debugging, and evaluation can also be connected.

AgentPortal 2.0 further solves the problem of unified use and management of agents, allowing enterprises to bring Agents built by different departments into a unified entrance, while also supporting centralized management of ADP and third-party agents.

Employees can directly raise business needs, and the system automatically selects suitable agents to handle them; enterprises can also distribute Agents by department and function, and uniformly manage listing, delisting, and usage permissions.

For complex projects that require long-term collaboration, multiple agents can form a team, with a project manager agent coordinating tasks; for businesses that directly serve consumers, multiple specialized Agents can also be gathered behind a unified service entrance.

In the past, enterprise software was usually procured and deployed by the IT department, which then required business departments to learn and use it. But after the spread of personal Agents, frontline employees may first discover that a certain AI working method is effective, and then push the enterprise to turn it into a formal application.

The two-way connection between ADP and WorkBuddy makes this path technically easier to realize.

Employees can first explore and validate working methods through WorkBuddy, then convert high-quality workflows into ADP agents and bring them under unified enterprise management; the Skills an enterprise already has can also return to employees' daily office entrance for continued use.

IV. From 1 project to 1,000 projects, can Tencent Cloud seize the ecological niche of enterprise AI?

If the previous three questions respectively correspond to how enterprise Agents enter business, generate business value, and achieve unified management, then there is an even more daunting challenge ahead: how to replicate one successful project across more enterprises?

The deeper enterprise Agents go into real business, the more they require professional engineering teams to enter customer sites. This is also why the FDE (Forward Deployed Engineer) model has regained attention.

FDE is a working method that has existed in the overseas enterprise software industry for many years. Engineers directly enter the customer's business environment, work with business teams to identify problems, connect systems, and develop solutions.

After the rise of AI, FDE began to play a more important role in enterprise Agent delivery.

In October 2026, Australian software leader Atlassian disclosed that its FDE team had already worked with more than 100 enterprise customers, driving more than 80 AI Agents into production environments.

FDE is valued precisely because real enterprises often cannot complete Agent deployment based on product manuals alone.

Business data may be distributed across different systems, and permission rules may not necessarily match actual workflows. Many problems only surface after engineers truly enter the business site.

But customized delivery also has a paradox: the more it relies on experts, the harder it is to reduce delivery costs.

In September 2026, Gartner offered a prediction: by 2028, 70% of enterprises may abandon Agents built by vendors' FDE teams, with major obstacles including increased costs and difficulties in independent maintenance.

This prediction also reminds the industry that if a project can only be kept running by the initial implementation team, it will be difficult to form a sustainable business model.

What enterprise Agents need is not just to complete a project, but also to distill the experience from the project so that subsequent customers can reuse it at lower cost.

Wu Yongjian told us that Tencent Cloud does not want to turn enterprise Agent delivery into a pure project or manpower outsourcing business. He used an analogy: in the past, doing enterprise Agent projects was somewhat like running a private kitchen, where every customer who came required the chef to customize according to specific needs. But if the business is to grow, it is necessary to gradually distill replicable dishes based on experience serving a large number of customers in the past.

"Only when replication efficiency improves do I think this business can run." Wu Yongjian said.

To this end, Tencent Cloud is trying to break through the dilemma of customized delivery with "FDE + ADP + scenario packages."

At this conference, Tencent Cloud launched FDE certification and further upgraded the ADP open interfaces and ADP CLI, promoting enterprises, developers, and ecosystem partners to participate in agent building and delivery.

In addition, Tencent Cloud has further distilled the project experience accumulated over the past few years into a delivery methodology that can be handed over to partners for use.

This FDE system covers five stages: the access stage, connecting systems and data and clarifying permission and security boundaries; the build stage, combining Claw Agent and Workflow to form business solutions; the validation stage, testing results and controlling risks through real scenarios; the promotion stage, driving Agents from pilots to scaled applications; and the governance stage, continuously managing knowledge, results, costs, and security.

Each stage is equipped with corresponding tasks, deliverables, and acceptance criteria, as well as templates, evaluation cases, and operation manuals, enabling partners to complete delivery according to a relatively unified process.

On this basis, Tencent Cloud has also summarized past project experience into 25 service scenario packages covering five major industries: retail, manufacturing, education, finance, and healthcare.

At the same time, Tencent Cloud also demonstrated progress in vertical applications such as intelligent customer service and visual agents, trying to distill more mature capabilities onto the ADP platform.

Currently, ADP has covered more than 1,000 public cloud application scenarios and more than 100 private cloud application scenarios, and has distilled replicable scenario packages in industries such as retail, finance, healthcare, and industrial.

There are many common needs behind these scenarios. For example, different enterprises all need to securely connect internal systems, manage Agent permissions, monitor operations, and adjust applications based on business feedback.

If these capabilities can be provided uniformly through the platform, engineers will not need to solve the same problems again in every project. But industry scenario packages do not mean that all enterprises can copy them directly. Therefore, FDE and internal enterprise teams still actively participate in adaptation.

In some scenarios with similar business processes and a high degree of standardization, this model has already shown the potential for scaled deployment.

For example, the hotel intelligent agent butler "Hua Xiao AI," built on ADP, covers 152 customer service sub-scenarios through 38 workflows and has already been deployed in more than 10,000 stores, automatically handling over 70% of high-frequency customer service inquiries. There are a large number of similar customer service needs across hotel stores, and the agent can reuse common workflows and then adjust them according to the specific needs of different stores.

Scenarios like this that can be repeatedly deployed across a large number of stores are more likely to gain economies of scale.

Therefore, Tencent Cloud's push to build the FDE ecosystem this time is also an attempt to use ADP and industry scenario packages so that engineers do not have to repeatedly solve the same problems in every project.

FDE solves the problem of getting into the field, platformization lowers the cost of repeated development, and industry scenarios and ecosystem partners support further replication. This is also where the moat for enterprise AI may gradually take shape.

For small and medium-sized To B AI companies, developing similar model applications or product features is not necessarily an insurmountable technical challenge. But entering enterprises' core business systems, gaining customer trust, accumulating cross-industry experience, and building a partner network covering development, delivery, and continuous operations all require longer-term investment.

Wu Yongjian also mentioned a similar judgment. In his view, models and the foundational technical capabilities around them are certainly important, but there are still a large number of "last mile" problems in landing enterprise Agents. How to solve these problems one by one, distill the solutions into products, and then deliver them at scale through partners is not low in difficulty.

He compared this process to going from cooking one dish well to being able to operate tens of thousands of stores. The former tests the capability of a single product, while the latter requires a complete system that can operate stably over the long term and be continuously replicated.

Tencent Cloud is using its existing enterprise ecosystem and industry deployment foundation to strive for a first-mover position. Whether this path can form a scale advantage will determine how deep the moat Tencent Cloud builds in the enterprise Agent market actually is.

In the past, enterprises bought software to make employees work more efficiently. Today, Agents are beginning to connect directly to business systems and execute tasks, and the role of people is gradually shifting to setting goals, reviewing decisions, and handling exceptions.

Tencent Cloud's ADP upgrade this time is precisely centered on these issues:

Downward, through secure connections, permission isolation, operation monitoring, and feedback optimization, Agents can enter business systems and run continuously;

Upward, through AgentPortal 2.0, intelligent agents are managed and distributed uniformly, and WorkBuddy is connected, allowing personal workflows to be distilled into services that enterprises can reuse;

Outward, with the help of industry scenario packages and the FDE ecosystem, it attempts to reduce the cost of repeated development and delivery for different enterprises.

This also reflects a change taking place in the enterprise Agent industry: the focus of competition is shifting from giving Agents more capabilities to how to organize these capabilities into enterprise services that can run over the long term.

Tencent Cloud's series of explorations around ADP this time are precisely trying to build the first moat for the scaled deployment of enterprise AI beyond model capabilities.