RM新时代有限公司

CloudWalk Technology Debuts at WAIC: Connecting AI Marketing and Physical AI with Trusted Agents

CloudWalk Technology Makes It Clear: Train Agents, Organize AI Teams, Enter Real Business

New Signals from WAIC

This year's WAIC sent a clear signal:

AI is stepping out of the chat box and entering the real world.

On one end, AI enters robots, factories, logistics, and public spaces, beginning to perceive environments, execute actions, and complete real-world tasks. Physical AI has become a focus of industry attention.

On the other end, AI enters the cross-border e-commerce operations, undertaking market insights, content generation, marketing, and business analysis, directly serving GMV growth.

The two directions may seem different, but they face the same set of issues:

Can AI work stably?
How to coordinate complex tasks?
How to verify results?
How to control risks?
How to continuously accumulate experience?

Large models make intelligence supply increasingly abundant; what is truly scarce begins to shift to human judgment and the organizational ability to stably coordinate multiple intelligent agents.

From July 19 to 20, Yang Hua, Senior Vice President and Board Secretary of CloudWalk Technology, attended the WAIC Financial Afternoon Tea and WAIC CONNECT events.

Commercialization ultimately depends on results

At the WAIC Financial Afternoon Tea on July 19, Yang Hua shared CloudWalk's observations on agent commercialization.

When enterprises focus on AI, they ultimately return to a few simple questions:

Has efficiency improved?
Have costs decreased?
Has revenue grown?
Has experience been retained?

For AI to enter enterprises, it needs to be integrated into high-frequency, closed-loop business processes with real feedback and auditability. A model completing a single task does not equate to stable productivity.

Models determine what AI can do; coordination capabilities determine whether AI can work long-term.

This is also the starting point for CloudWalk's AI marketing team.

Cross-border e-commerce has long faced issues such as a large number of SKUs, complex marketing chains, scarce operational talent, and difficulty in accumulating experience. CloudWalk's AI marketing team consists of human experts and agent digital employees, continuously working on listing optimization, content generation, advertising, competitive analysis, and business review.

Agents can simultaneously handle more product and advertising data, while human experts are responsible for goals, budgets, branding, and key judgments.

What AI undertakes here is a continuous operational job.

As agents increase, who will train and manage them?

At the WAIC CONNECT event on July 20, Yang Hua further shared CloudWalk's views on AI organizational forms.

Large models already possess increasingly strong understanding, reasoning, generation, and multimodal capabilities, but a smart brain cannot automatically become a qualified employee.

It needs to learn job responsibilities, business processes, professional skills, and safety rules; undergo training, testing, and assessment; and after entering work, it requires scheduling, management, acceptance, and continuous review.

When an enterprise simultaneously has ten, fifty, or even hundreds of agents, who assigns tasks, organizes coordination, controls permissions, checks results, and accumulates experience becomes a new management challenge.

CloudWalk is building a complete agent training and operation system.

In this system:

The Xingyun Agent Operating System is a "vocational school."

It is responsible for organizing training, arranging collaboration, managing processes, and accepting results, enabling agents to grow from having general intelligence to becoming professional roles that can enter business processes.

Mofang Agent Lab is a continuously updated "textbook."

It consolidates job experience, industry knowledge, business rules, and professional methods into Roles, Agents, Skills, and task templates, providing agents with capabilities they can learn, invoke, and iterate.

The Controlled Training Ground and Embodied Intelligence Training Ground provide training, testing, and real-world environments.

Agents can repeatedly learn, execute, correct, and verify in these environments, and after passing safety and quality checks, they enter real business.

Xingyun is responsible for training and management, Mofang provides knowledge and skills, and the training ground tests whether they can truly work.

The same batch of trusted agents enters two types of real worlds

Trained and verified trusted agents can enter digital business or physical spaces.

In cross-border e-commerce scenarios, they form an AI marketing team, undertaking market analysis, content production, advertising, operational optimization, and growth review.

Every agent has a clear role, every operation has a process record, and every round of marketing results flows back to form the basis for the next round of optimization.

Ensure every SKU is meticulously managed, and every marketing effort leaves reusable insights.

In physical AI scenarios, trusted agents can connect robots, devices, and task workflows, participating in real-world tasks such as industrial manufacturing, logistics, and inspection.

The physical world is full of continuous changes, environmental disturbances, and safety constraints. Robots need to perform actions, but also understand tasks, handle anomalies, pass verification, and collaborate stably in complex scenarios.

CloudWalk's deployment of embodied AI training grounds is precisely to allow agents to undergo thorough training before entering real spaces and accumulate experience during continuous operation.

The competition in physical AI is shifting from 'whether it can move' to 'whether it can work stably.'

The digital world and the physical world have different requirements for agent capabilities, but the underlying logic is the same:

Both require specialized training, human judgment, controllable processes, and continuous learning.

From AI Infrastructure to AI Teams

Around trusted agents, CloudWalk is forming a comprehensive layout.

The bottom layer consists of the CongRong multimodal large model, Yunqi Model Hub, trusted data space, training and inference platform, controlled training ground, and embodied AI training ground, providing agents with model, data, training, and verification capabilities.

The middle layer is the Xingyun Agent Operating System and Mofang Agent Lab, responsible for agent cultivation, organization, scheduling, and continuous evolution.

The top layer consists of AI teams entering real-world scenarios, including AI content production studios, AI marketing teams, AI functional teams, AI R&D teams, and future professional teams for more industries.

Infrastructure provides intelligence, training systems cultivate professional capabilities, and AI teams are responsible for delivering results.

AI's Next Stop: The Real World

From cross-border e-commerce marketing to robots and industrial sites, CloudWalk conveyed a clear theme at this year's WAIC:

AI is leaving the demo environment and entering real business and real spaces.

After entering the real world, model capability is just the starting point. Training, judgment, collaboration, governance, and continuous delivery will determine how far AI can go.

CloudWalk hopes to cultivate more verified trusted agents through the Xingyun Agent Operating System, Mofang Agent Lab, and two types of training grounds, enabling them to form professional AI teams that enter business operations and the physical world.

AI now has a brain; the next step is to receive education, complete training, and join organizations.

What CloudWalk aims to do is to put trusted agents to work for real.