At the 2026 Tencent Cloud AI Industry Application Conference, Tencent Cloud launched data infrastructure products such as DataBuddy and CFS Turbo, along with scenario-specific Agents like MAGIC AI Marketing Cloud and DatabaseClaw, addressing the three major challenges enterprises face in deploying Agents: data silos preventing usage, insufficient security and reliability causing hesitation, and excessive costs making adoption unfeasible.

Author: Lian Ran

Source: GeekPark

At the 2026 Tencent Cloud AI Industry Applications Conference on June 5, more than 20 AI-native products were unveiled collectively. From marketing and e-commerce to office productivity and risk control, as well as content creation and development collaboration, Agent technology appeared in nearly every core enterprise scenario.

Public attention has largely focused on frontend applications such as WorkBuddy, CodeBuddy, and ima, which can write code, generate PPTs, and assist with office tasks—smart enough to demonstrate the growing potential of agents entering real-world work environments. However, if we shift our perspective to industrial practice, we find that numerous practical challenges still need to be addressed to shape the future of the agent industry.

Over the past year, agents have progressed from proof of concept to real-world implementation. An increasing number of enterprises are beginning to integrate agents into their business processes, and some have even established dedicated agent teams. However, widespread challenges remain: difficulty in adoption, lack of trust in their use, and high costs.

It’s hard to use because enterprise knowledge, business processes, and data systems have long been siloed—agents know how to complete tasks but can’t access the required information. It’s hard to trust because agents lack long-term memory and stability, leaving persistent risks around data security, access control, and compliance. And it’s too expensive because token costs, deployment complexity, and operational barriers still limit scalable adoption.

These issues do not stem from the model's capabilities, but from outside the model.

For this reason, Tencent Cloud has not focused on a single Agent product but instead aims to build a comprehensive enterprise-grade Agent infrastructure centered on three capabilities: scenario connectivity, engineering mastery, and model-driven power.

Behind this lies a fundamental shift in the competitive logic of the entire Agent industry. Previously, the industry competed on model capabilities; now, the competition is shifting toward who can truly integrate Agents into enterprises, embed them into workflows, and ultimately drive productivity.

01 Integrate data and processes to give Agent real tasks to accomplish

After an agent enters a company, the first issue that often surfaces is not model capability, but business capability.

It knows how to create marketing plans but doesn’t understand the company’s products or customers; it knows how to complete approval processes but can’t see the internal rules of the organization; it knows how to analyze problems but can’t access real data from business systems.

This is the first core obstacle to implementing Agent: scenario disconnection and data silos.

A company’s operations are complex and fragmented; a general-purpose agent cannot cover all scenarios. Meanwhile, enterprise data is scattered across dozens of different systems—such as ERP, CRM, and databases—each with its own interfaces, permissions, and data formats. Agents often know “what to do,” but cannot access the “data needed,” and are unable to integrate into existing enterprise workflows.

Tencent Cloud's approach is to first establish a solid data foundation, then integrate Agent capabilities into every core business scenario.

At the Industry Application Conference, Tencent Cloud launched two products, DataBuddy and CFS Turbo, to fundamentally solve the issues of data being unusable and unmovable.

DataBuddy is an AI-powered agent workspace designed for the entire big data lifecycle, fully automating data engineering, data governance, and data analysis—tasks that previously relied heavily on specialized personnel. ETL scripts that once took data engineers days to write can now be automatically generated through natural language. Previously, only 30% of core tables could be configured with quality monitoring; with DataBuddy, the entire data warehouse is automatically scanned, intelligently identifying sensitive fields and recommending monitoring rules, achieving 100% coverage.

It has transformed the traditional paradigm of "humans operating tools" into "AI works + humans oversee," turning data from a "technical asset" accessible only to engineers into a "means of production" that all agents can utilize.

DataBuddy solves the issue of whether data is easy to use, while CFS Turbo addresses whether data transfers quickly.

In traditional AI workflows, data must be repeatedly transferred between object storage and high-performance storage, which is not only time-consuming and labor-intensive but also prone to data inconsistencies.

CFS Turbo, a high-performance parallel file storage system, has for the first time achieved real-time strong consistency between file semantics and S3 object semantics, enabling enterprises to complete the entire data workflow—from ingestion and cleaning to training and inference—on a single unified platform without any data migration. Additionally, its metadata search engine and natural language Agent retrieval capabilities have improved search efficiency for petabyte-scale data by a thousandfold, allowing Agents to locate any required data instantly, just like dedicated employees.

With a solid data foundation, seamlessly integrate Agent capabilities into all of the enterprise's business scenarios.

MAGIC AI Native Marketing Cloud by Tencent for marketing, and the Brand Omnichannel Operations Cloud Mall for transactions—both products are built on an “Agent Collaboration” architecture, driving enterprises from a “human-configured system” to an “Agent-organized system.” Leveraging a unified data foundation to connect “public-domain customer acquisition ↔ private-domain operations” and “marketing conversion ↔ transaction repurchase,” they empower businesses to build an omnichannel intelligent growth loop.

From Tencent Meeting AI for office workflows, to Tianyu Risk Control Agent for risk management, WAND for audiovisual AI in content creation and processing, Live-action Drama Agent, and DatabaseClaw database Agent for operations and maintenance...

In operational scenarios that heavily rely on professional expertise and real-time data, DatabaseClaw further demonstrates how an Agent can achieve a true execution loop based on a data foundation.

DatabaseClaw, a production-grade database AI Agent developed by Tencent Cloud based on over 100,000 real internal DBA tickets, can directly connect to databases to access data and business metrics, completing the entire process—from monitoring and analysis to execution—without requiring manual log exports or metric uploads.

In the past, after a database anomaly occurred, DBAs typically had to manually review monitoring metrics, examine SQL logs, analyze traffic patterns, and then use their experience to determine the root cause.

Now, when the system triggers a slow query alert, DatabaseClaw can automatically retrieve performance trends, SQL execution logs, and business traffic data from the past seven days to identify the root cause and generate corresponding optimization recommendations. After human confirmation, the Agent can automatically perform operations such as index optimization and parameter tuning, reducing what used to take hours of troubleshooting to just minutes.

More importantly, it enables database operations with natural language interaction, allowing even non-professional DBAs to perform database inspections, performance diagnostics, and troubleshooting through conversation.

This "data foundation + scenario integration" model is fundamentally about having the Agent adapt to employees' existing workflows, not the other way around. Employees can access AI capabilities directly within the marketing tools, office software, and operations platforms they use daily—without needing to learn new systems.

This may be the key to enabling agents to truly gain widespread adoption in enterprises.

02 To enter the core business, being smart isn't enough.

Having addressed the question of "whether there's work to be done," the next challenge is an even more practical one: whether enterprises dare to entrust critical tasks to Agents. This remains one of the biggest hurdles to the real-world adoption of enterprise-grade Agents.

Over the past year, many companies have begun experimenting with integrating agents into real business scenarios, but most applications still remain at an auxiliary level: looking up information, writing weekly reports, generating copy, and organizing meeting minutes. When it comes to core functions such as decision-making, operations, and risk control, companies tend to become especially cautious.

The reason is not complicated. Current agents are still not very reliable. They may "forget," losing context midway through a conversation; they might also confidently provide incorrect advice or faulty decision-making recommendations. More importantly, entrusting core data and business operations to a black-box AI could lead to catastrophic consequences if data leakage or operational errors occur.

Once the agent begins accessing enterprise knowledge, business processes, and core data, the issue is no longer model capability, but reliability, controllability, and security.

The suite of products launched by Tencent Cloud is centered on the core principles of "reliability" and "security," building a comprehensive engineering capability system that transforms agents from merely "usable" to "trustworthy."

The first issue to address is the Agent's "amnesia" and "lack of understanding of businesses." The greatest value of a business Agent lies in its ability to understand and retain enterprise-specific knowledge and expertise.

Tencent Lexiang has redefined the form of knowledge bases.

Traditional knowledge bases served employees, while in the Agent era, knowledge bases begin serving Agents themselves. Documents, cases, experiences, and processes are no longer merely read—they can be retrieved, invoked, and further transformed into capabilities that Agents need to execute tasks.

In a sense, enterprises are undergoing a restructuring of their knowledge systems. In the past, knowledge was embedded in people; in the future, it must be embedded in systems that agents can understand and access. Companies can store all their documents, manuals, case studies, and expertise in Tencent Lexiang, where agents can not only retrieve information accurately through natural language but also encapsulate experts’ methodologies into Skills, enabling scalable reuse of capabilities.

But knowledge is only the first step. More important than knowledge is memory.

Most agents today still operate at the stage where they end once a task is completed. After a task ends, the context disappears, experiences cannot be retained, and organizational knowledge cannot accumulate. This is why many enterprise agents appear intelligent but never truly grow.

Tencent Cloud Database's Agent Memory service aims to solve exactly this problem.

It establishes a comprehensive lifecycle management system spanning short-term memory, long-term memory, and team memory. At the short-term memory level, through context compression technology, it has developed proprietary symbolic compression and context offloading capabilities, improving agent task success rates by 30% in long-task scenarios while reducing token costs by 30% to 60%.

At the long-term memory level, OpenClaw’s long-term memory capability has been significantly enhanced through a four-tier progressive memory retrieval system. On the PersonaMem benchmark dataset, we improved OpenClaw’s native memory evaluation score from 48% to 76%.

At the team memory level, a three-tier permission system has been established—comprising individual private domains, departmental collaboration zones, and an organizational global library—to organize team context scattered across conversations, tasks, documents, and workflows into a shared memory layer reusable by multiple agents.

More importantly, it automatically transforms employees' tacit knowledge into actionable Skills, truly achieving "knowledge remains when people leave." Once the Agent possesses the team's memory, it begins to exhibit the characteristics of a "digital employee." Every use, collaboration, and decision made by the company has the potential to be converted into new organizational assets.

However, when agents truly enter core business operations, security issues become a new challenge.

In the past, security was primarily an IT systems issue; in the Agent era, security is becoming an AI governance issue.

The agent has the ability to invoke tools, access data, and perform tasks, which means it has substantial access to enterprise data, posing potential data security risks. For industries such as finance, government, and healthcare, these risks are particularly sensitive.

The AICC Trusted Cluster is dedicated to enabling trustworthy, controllable, and verifiable large model inference capabilities, ensuring that enterprises' agents do not leak any data when invoking LLMs.

Built on a hardware root of trust, AICC creates an end-to-end encrypted inference environment where all data remains encrypted throughout its entire journey and never appears in plaintext at any stage. It also provides provable security mechanisms, enabling enterprises not only to know their data is secure but also to demonstrate this security to customers and regulators. In terms of usability, AICC enables rapid deployment of inference nodes in just 30 seconds, supports all major models, and significantly reduces inference costs through multi-level KVCache optimization.

The AICC Trusted Cluster addresses the two major concerns of enterprises—data leakage and compliance verification—enabling agents to finally enter high-security core scenarios such as finance, government services, and healthcare.

As companies begin to demand that AI not only perform tasks but also explain how it works and prove its safety, competition among agents is shifting from model capability to engineering capability.

03 From experimental project to foundational capability

After the Agent solves the issues of "usability" and "trustworthiness," the final hurdle is "affordability." This is the key to whether the Agent can achieve scalable deployment.

Today, many enterprises' agent projects are facing issues with uncontrolled costs.

For a medium-sized business, monthly token costs can reach hundreds of thousands or even millions; when combined with expenses for server deployment, operations and maintenance, and model optimization, agents have become a "luxury" accessible only to leading enterprises. Meanwhile, individuals and small-to-medium businesses face extremely high technical barriers when attempting to deploy their own agents.

Powered by Tencent Cloud’s model capabilities, technological innovation, and engineering optimizations significantly reduce the cost of deploying and using agents, transforming them from a "luxury" into an accessible tool available to all businesses and individuals.

On the deployment barrier, Tencent Cloud has launched an all-in-one solution ranging from individuals to enterprises. For individual users and developers, LightHouse provides one-click deployment of cloud-based Agents, enabling users to have a dedicated AI assistant available 24/7 without requiring server administration knowledge. It also features functionalities such as "Crab Hospital" and Cloud Agent Chat, addressing the pain points of difficult deployment and unstable operation of open-source Agents. To date, it has built a developer community of nearly 100,000 members.

For enterprise users, ClawPro’s Enterprise Agent Control Console enables unified management of agents, permission auditing, cost monitoring, and skill library development. Enterprises can assign digital employees to staff with a single click, precisely control each agent’s permissions and costs, and overcome the management challenges of scaling agent deployment.

On inference costs, Tencent Cloud's Large Model Service Platform, TokenHub, achieves极致 utilization of computing power through a series of technological innovations.

It supports Hunyuan and all major third-party models, automatically performing intelligent multi-model routing based on task complexity and cost requirements, ensuring every token is used efficiently. Leveraging technologies such as Tide Scheduling and FlexKV distributed caching, TokenHub has increased overall compute utilization by 40% and boosted cache hit rates to 85%, effectively reducing Agent inference costs.

This "low barrier + low cost" model makes the scalable deployment of agents possible.

Internal practices at Tencent Cloud have demonstrated this: CodeBuddy covers over 95% of Tencent’s engineers and reduces overall coding time by 40%; WorkBuddy enables human-AI collaborative development, allowing small teams of just a few people to complete an initial version in two days while maintaining a two-day release cycle.

When agents begin to leave the lab and enter enterprise production workflows, the factors determining their value are no longer just the models themselves, but the entire infrastructure supporting their operation.

Scenario connectivity gives agents tasks to perform, engineering control ensures agents are reliable and secure, and model-driven power makes agents affordable and scalable. These three capabilities form a complete closed loop for enterprise-grade agent deployment.

In the competition within the Agent industry, whoever builds the most comprehensive, user-friendly, and secure Agent execution environment will gain a competitive advantage.

This also signals an important shift from Tencent Cloud’s recent release—that the Agent industry is transitioning from a capability validation phase to a stage of large-scale deployment.

The future competitiveness of enterprises will depend on their ability to integrate agents into organizational operations at sufficiently low cost and high efficiency.