The era of high-code development has arrived. Alibaba Cloud's Bailian aims to integrate Agents into real business operations.

“In the era of AI, the development of intelligent agents is an important development paradigm today.” At the 2025 Yunqi Conference held on September 24th, Zhou Jingren, the Chief Technology Officer of Alibaba Cloud Intelligence Group, mentioned “Agent” 38 times and “intelligent agent” 33 times.

Undoubtedly, Agent is becoming an inevitable path for enterprises to pursue efficiency improvement. In May this year, after PwC surveyed 300 executives globally, it was found that 79% of the surveyed companies have already applied AI Agents in some of their businesses; among them, 66% reported increased productivity, 57% saw cost reduction, 55% felt faster decision - making efficiency, and 54% improved customer experience.

Currently, Agent platforms have become a hotly contested territory for large Internet companies. Overseas, LangChain/LangGraph, Microsoft AutoGen, and Google ADK have all been upgraded. In China, Tencent, Huawei, and Alibaba Cloud's Bailian launched and upgraded their Agent intelligent agent platforms in September 2025.

However, beneath the industry's enthusiasm, the real - world implementation experience may still need further examination. For example, most of the low - code Agent platforms that have emerged in the past two years rely on the “pre - defined orchestration” logic. They are suitable for simple Q&A or process automation, but often struggle when it comes to cross - system calls, long - chain tasks, or when multiple rounds of reflection and autonomous decision - making are required. More importantly, what enterprises really care about is not “whether an Agent can be developed,” but “whether the Agent can operate stably in the business system,” and ideally, achieve “scale and commercialization.”

At this year's Yunqi Conference, Alibaba Cloud provided an answer. Alibaba Cloud's Bailian platform newly launched the ModelStudio - ADK high - code framework, directly addressing these pain points of enterprises. It not only enables enterprises to efficiently develop Agents with autonomous planning and cyclic execution capabilities but also adds a matrix of seven enterprise - level capabilities, including tool invocation, memory access, dynamic reasoning scheduling, sandbox environment, full - link observability, and even a payment channel jointly launched with Alipay.

This choice also gives Alibaba Cloud more depth in the competition. While competitors are still emphasizing “low threshold and quick start,” Bailian is trying to add “high - code and strong capabilities.” Alibaba Cloud's Bailian runs on both low - code and high - code tracks, thus creating production - level Agents that can truly be implemented for enterprises.

01

Bailian, a Step Towards ASI

Wu Yongming, the CEO of Alibaba Group, judged at the 2025 Yunqi Conference: “Artificial General Intelligence (AGI) is not the end of AI development but a brand - new starting point. AI will not stop at AGI; it will move towards Artificial Superintelligence (ASI), which surpasses humans and can self - iterate and evolve.”

Agent is a step towards ASI and one of the ways to enable large models to be truly implemented in industries and release practical value.

The positioning of Alibaba Cloud's Bailian platform is closely related to Alibaba's judgment on the future evolution of AI. It provides the “soil” for enterprises to truly grow AI Agents. Here, developers can obtain the full set of capabilities required for intelligent agents at once: from model invocation, framework construction, to resource scheduling and compliance operation and maintenance, all are systematically packaged together.

At the 2025 Yunqi Conference, Alibaba Cloud's Bailian platform was comprehensively upgraded. As a key bridge for Alibaba's AI to reach all industries, Bailian extended further to the development and implementation of Agents on the basis of its original model services.

What supports all this is Alibaba's complete industrial chain from the bottom up: at the top are Tongyi series of large models, in the middle is Bailian's enterprise - level development and operation platform, and at the bottom are cloud computing infrastructures such as computing power, storage, and databases. This chain enables enterprises' Agents today to be not only quickly developed and stably operated but also continuously iterated with technological evolution, laying a solid foundation for future intelligent development at a higher stage.

Based on the complete technology stack, in the past year and more, more than 200,000 developers on Alibaba Cloud's Bailian platform have developed over 800,000 Agents, and these rich experiences have been precipitated into a complete system covering frameworks, models, and components:

At the framework level, Alibaba Cloud's Bailian supports parallel development and phased implementation: the low - code ADP allows enterprises to quickly verify prototypes like assembling Lego bricks and quickly translate ideas into specific scenarios; the high - code ADK opens up underlying interfaces, supporting in - depth customization and large - scale deployment of complex businesses. In the early stage, ADP can be used for rapid trial - and - error to find effective application scenarios; once verified, it can seamlessly switch to ADK to support in - depth customization and large - scale operation.

At the model level, Bailian continuously updates the flagship models of the Tongyi Qianwen family. The latest Qwen3 series has stronger reasoning capabilities, which can significantly improve the autonomous planning and decision - making effects of Agents - the reasoning performance is improved by 50%, and the decision - making success rate reaches 90%. At the same time, the Bailian platform also opens up the invocation interfaces of more than 200 leading models in the industry, from Qwen, Wan to DeepSeek. Enterprises and developers can access them with just one click.

At the component level, Bailian builds seven enterprise - level capabilities: the MCP Server for tool connection, the RAG Server for multi - modal data fusion, the system - level sandbox Sandbox, the intelligent memory Memory Server, the full - link observability system, dynamic reasoning scheduling, and the Pay Server jointly launched with Alipay.

Among them, the Pay Server is particularly worthy of attention - it is the industry's first professional commercial payment channel for enterprise - level Agents, which means that Agents can not only be developed and run but also directly enter the commercialization stage.

With the continuous improvement of model capabilities and the explosion of Agent applications, Alibaba Cloud's Bailian platform has also taken the lead in presenting an outstanding report card. In the past year alone, the monthly model invocation volume of the platform has increased by 15 times, meeting the AI application development needs of enterprises and developers in all industries.

02

Implementing in Real Business Scenarios: ModelStudio - ADK

“We hope to enable the model to understand a complex task, decompose it, execute it respectively, plan, make decisions, and continuously correct various problems encountered in the execution process, so as to better solve practical problems in our business scenarios.” At this year's Yunqi Conference, Zhou Jingren talked about his vision of the future of Agents.

Alibaba Cloud's newly launched Agent development framework, ModelStudio - ADK, is part of this vision. The goal of this framework is to break through the pre - defined orchestration method and help enterprises efficiently develop Agents with autonomous decision - making, multi - round reflection, and cyclic execution capabilities.

Low - code/zero - code Agent development platforms have been the mainstream products in the past two years. The execution mode of the Agents generated by them highly depends on “pre - defined orchestration,” that is, developers need to pre - define when to call what tools and follow what paths. Agents execute tasks according to a “scripted” process.

The advantage of this method is that as a developer, you don't need to know too much code. You can also launch an Agent that can answer questions and execute fixed processes by dragging and dropping a few modules, and the development speed is very fast.

However, this development method can only handle simple and well - defined tasks. Once it comes to complex businesses, problems will arise: in complex business scenarios, the Agents developed in this way cannot “reflect” or “adjust on the spot” like humans and can only follow the set rules; an enterprise often has dozens or hundreds of IT systems: CRM, ERP, payment, risk control, supply chain, but pre - defined orchestration requires writing all interfaces and logic in advance, and the maintenance cost is extremely high. Any slight change requires rewriting from scratch.

While lowering the threshold, it also brings new limitations: lack of autonomy, complexity in cross - system operations, insufficient stability, and a gap in commercialization... The industry is in an awkward situation: everyone can create Agents, but few can make them truly run in a production environment.

This is why Alibaba Cloud's ModelStudio - ADK focuses on the high - code framework. As a development framework specifically designed for enterprise - level intelligent agents, it is built on Alibaba's open - source AgentScope. The core breakthrough lies in: no longer relying on traditional “pre - defined orchestration,” but supporting developers to use a more flexible way to enable Agents to have the capabilities of autonomous decision - making, multi - round reflection, and cyclic execution.

It is reported that ModelStudio - ADK can develop applications such as in - depth research, hardware proxy intelligent agents, and complex retrieval intelligent agents, and it has established the complete closed - loop from model invocation to deployment and operation, and then to observability and commercialization.

“It defines various core capabilities and components of Agents to meet the construction logic under complex task scheduling through a code - based approach combined with the capabilities of models.” Xu Zhiyuan, a senior product expert at Alibaba Cloud's Bailian, summarized it in this way.

In the development stage, ADK emphasizes the native code - based experience, introducing autonomous decision - making, dynamic reflection, and cyclic execution logic, enabling Agents to truly handle complex scenarios instead of just executing preset processes. In conjunction with this, the platform opens up the invocation interfaces of more than 200 mainstream models, supporting one - click access in the cloud and being compatible with enterprise - level local deployment, helping developers freely switch between “performance and controllability.”

Centering on the capabilities required for business implementation, Alibaba Cloud packages key components such as tool invocation, knowledge base retrieval, memory management, and sandbox operating environment into standardized services. Developers can directly access them through APIs. This means that enterprises can quickly build Agents connected to business systems without repeatedly building underlying facilities and can call the security - isolated sandbox environment to run code or control devices when needed.

In the operation stage, ADK provides flexible deployment methods: one line of code can deploy Agents to the cloud or locally, while ensuring security isolation and elastic scaling. In the operation and maintenance and optimization stage, the platform has a built - in complete observability system, with real - time monitoring of requests, effects, and performance, helping enterprises manage the behavior and efficiency of Agents like managing software engineering.

More importantly, ADK is compatible with multiple mainstream protocols, supporting synchronous and asynchronous calls. Enterprises can freely combine multiple Agents into a collaborative system to explore the application mode of “multi - intelligent agents.” To lower the learning threshold, Alibaba Cloud has also open - sourced sample projects such as DeepResearch, Agentic - RAG, and Computer Use. Developers can directly experience or conduct secondary development to accelerate business trials.

Overall, this system solves the problem in Agent development in the past of “being able to create demos but difficult to implement on a large scale”: it not only ensures flexibility and stability in complex scenarios but also supplements the controllability, observability, and commercialization capabilities that enterprises care most about.

In contrast, the low - code platform is more like allowing developers to build flowcharts, while ModelStudio - ADK gives them real programming languages and execution engines. For enterprises that need to handle complex scenarios such as financial risk control, medical decision - making, and supply chain scheduling, this difference is crucial.

As Zhou Jingren said, Alibaba Cloud hopes that “our developers can quickly develop intelligent agents on the Bailian Agent platform.”

Currently, Alibaba Cloud's Bailian has launched a batch of ready - to - use intelligent agent applications such as DeepResearch, Agentic - RAG, and Computer Use based on ADK. Users can experience them online or download the source code for secondary development. These practices are transforming “whether it can be developed” into “whether it can run stably and bring value.”

03

Systematic Layout, Connecting the Whole Link

Agents are going through a clear evolution path: from technical experiments in the laboratory to large - scale production in enterprises. In the past two years, almost all cloud providers have been increasing their investment in the Agent track, but the paths are different.

Compared with competitors who emphasize single - point capabilities more, Alibaba Cloud's approach is more “integrated.” It doesn't just provide a model, doesn't just launch a low - code tool, nor does it just rent a server. Instead, through the systematic layout of “platform + framework + components,” it connects the whole link of enterprises from development to deployment, from operation to commercialization.

The upgrade of Bailian is a concentrated manifestation of this logic: the combination of “model + components + applications” not only ensures the continuous iteration of underlying reasoning and planning capabilities but also lowers the development threshold with standardized components and ready - to - use examples. This enables Agents to truly become productivity tools and be implemented on a large scale.

Alibaba Cloud's Bailian platform is playing a role in all industries. MYbank has developed a loan review application on ModelStudio - ADK. It can process 26 types of vouchers such as contracts, invoices, and business licenses simultaneously and identify more than 400 fine - grained objects such as store facades, restaurant kitchens, dining areas, and shelf products, with an identification accuracy rate of over 95%. Thanks to this ability, the original 3 - hour manual operation has now been compressed to less than 5 minutes. This is a manifestation of the value of the high - code framework in complex scenarios.

Sam Altman, the CEO of OpenAI, wrote at the beginning of this year: “We believe that, in 2025, we may see the first AI agents ‘join the workforce’ and materially change the output of companies.”

With the help of a full - stack AI service provider like Alibaba Cloud, this is becoming a reality.