From MCP to Super Agent: Who will be eliminated by this AI productivity revolution?
Updated on:June-26th-2025
Recommendation
The AI revolution is coming. How can enterprises stand out?
Core content:
1. The AI competition landscape has changed dramatically. MCP protocol and super agent have become new focuses.
2. Baidu Smart Cloud Qianfan large model platform has been completely upgraded to open up the "last mile" of application.
3. Model performance has become the key to enterprise competition. Qianfan platform helps enterprises build exclusive engines.
Yang Fangxian
Founder of 53A/Most Valuable Expert of Tencent Cloud (TVP)
The competition landscape of large models has changed: the underlying technology advantage is no longer the key to victory, and the ability to build an ecosystem around the MCP protocol and super agent is becoming the core arena for unleashing the value of applications.Just last Friday at the Create 2025 Baidu AI Developer Conference, Baidu Smart Cloud Qianfan Large Model Platform unveiled its latest strategy: integrating hundreds of industry-renowned models developed by Baidu itself and third parties, providing comprehensive model customization and fine-tuning tool chains, launching an enterprise-level Agent development tool chain, and fully embracing the MCP protocol.Baidu Smart Cloud has built a development platform that covers the entire life cycle of large models and in-depth enterprise applications, focusing on model services, model development, and application development . Baidu founder, chairman and CEO Robin Li has pointed out the key direction of current AI development: "There will be many models, but what really rules the world is applications. Applications are king ."At a time when AI technology is advancing rapidly, the rise of the MCP protocol and super agents is setting off a subversive productivity revolution . As the "universal plug" of the AI world, MCP breaks the connection barrier between models and external tools, while super agents, with their powerful autonomous decision-making and task execution capabilities, integrate multiple resources with the help of MCP, pushing AI application efficiency to a new level.The competition for big models has entered deep waters, and the focus of the competition has shifted from the "intelligence ceiling" of AI to the "demand adaptation" of applications . A productivity revolution related to survival and elimination is taking place.One-stop intelligent infrastructure layoutOpen up the "last mile" of large model application
The intelligent ecosystem has entered a period of rapid expansion, and the challenges it brings to enterprises in building applications are becoming increasingly diverse. However, the core needs of enterprises remain the same: "precise adaptation" and "robust implementation" .It needs to meet the refined needs of vertical fields and achieve functional integration and efficiency optimization through a combination of multiple tools. The computing performance of model deployment, the stability and scalability of application development, and data security have become hard indicators throughout the entire process, directly affecting the actual effect and long-term value of application implementation.These links are closely related and indispensable. The role played by Qianfan Big Model Platform is to provide full-link capabilities to empower enterprises, and to accelerate and simplify complex processes for enterprises while accurately anchoring enterprise needs.As Shen Dou, executive vice president of Baidu Group and president of Baidu Intelligent Cloud Business Group, said: "What Baidu Intelligent Cloud has been doing is to build a set of intelligent infrastructure to minimize the trial and error costs of customers and partners and make innovation easier."This time, the intelligent infrastructure built by Baidu Smart Cloud has achieved system-level capability upgrades from computing power, models to applications , and has been running stably in multiple enterprise scenarios.Looking at the big model industry as a whole, companies' exploration of scenario-based application development is being linked with big model platforms that have full-stack technology support, forming the core dual engines that drive the prosperity of the application ecosystem. There is no doubt that promoting useful applications to take root in the core business scenarios of enterprises has become the watershed of this technological competition.Model side: precise customization and efficient developmentUnlock the enterprise-specific engine
For enterprises, the quality of model performance is directly mapped to the quality of application experience, becoming a key variable affecting user retention and business value release. This is why Shen Dou said that "the model effect directly determines the application experience."At the model level, Qianfan large model platform targets two things: model calling and development .Model calling means that enterprises have enough models to choose from. Qianfan Big Model Platform has integrated more than 100 models , including the latest released Wenxin Big Model 4.5 Turbo and Deep Thinking Model X1 Turbo and other Baidu's self-developed Wenxin series models, as well as third-party language and multimodal models such as DeepSeek, Llama, Tongyi Qianwen, and Shengshu Technology Vidu.Model development is aimed at the process of enterprises' demand for models from "general adaptation" to "exclusive customization". In the process of large-scale model implementation, general models can no longer meet the differentiated needs of enterprises, and the market demand for customized specialized models continues to rise , becoming a new rigid demand for intelligent transformation.The full-stack tools provided by Qianfan Big Model Platform give enterprises the ability to customize models right out of the box. Specifically, from data management, model fine-tuning, model evaluation and optimization to prediction service deployment, Qianfan Big Model Platform runs through the entire model development process with an integrated tool chain , which can support the customization and fine-tuning of various models such as deep thinking and multi-modality.Specifically for different industries, recruitment is actually a very typical scenario, which needs to meet the requirements of high real-time, multi-modal processing, high concurrency stability, low cost, etc. Although the existing industry deep thinking model can improve its service efficiency and quality to a certain extent, the actual operation will make the problems of reasoning delay and high cost prominent.Therefore, after realizing that the general deep thinking model could not meet the needs of users in high-concurrency scenarios, Zhaopin.com, a leading domestic recruitment platform, performed model distillation based on the Qianfan big model platform tool chain, achieving the effect of maintaining model performance and reducing costs to 1/3 of the original .It uses DeepSeek-R1 with hundreds of billions of parameters as the "teacher model" and ERNIE Speed with tens of billions of parameters as the "student model". It obtains the data set by distilling DeepSeek-R1, and then fine-tunes the student model based on the data set.In 2025, large-scale model iterations will enter the "fast lane". Faced with the greatly shortened model update cycle, the industry's doubts about whether the application value will be diluted by the rapid iteration wave are deepening.Practical cases represented by Zhilian Recruitment just answer a major concern in the current industry: high-quality tools are like precisely meshed "adaptive gears", which can closely connect the ever-changing technology with corporate needs, allowing the application value to continue to accumulate in the deep collision of business and technology.Just as Li Yanhong emphasized, "As long as you find the right scenario, choose the right basic model, and know how to adjust the model, the application you make will never become outdated . "Application side: Accelerating the prosperity of Agent ecosystemA new paradigm for enterprise-level AI collaboration
Although the model layer has built a strong capability foundation, there is still room for improvement before it can truly perform tasks efficiently. The Agent ecosystem layout , which is considered to be the best carrier for the application of large models , has become the key to bridging the gap between large model capabilities and enterprise business.Qianfan Big Model Platform released the Agent framework in September 2023, upgraded the enterprise-level Agent in September 2024, and took the lead in China to achieve full ecosystem compatibility with the MCP protocol in March this year. Today, its integrated Agent-related capabilities are becoming more and more complete.Nowadays, there is an increasing demand for Agent to be customized for enterprises in practical applications .This is because the Agent needs to combine the company's private domain data and corporate knowledge base to perform tasks based on industry experience and personalized understanding of business scenarios; in order for the Agent to follow corporate regulations, customized training of corporate standard SOP (standard operating procedures) is required; the Agent needs cross-platform and cross-tool calling capabilities, and integrates various resources and technical components; finally, it must be safe and stable, so that the Agent can achieve full process traceability and dynamic regulation.All of these can be easily solved on the Qianfan large model platform. This time, the Qianfan large model platform has once again upgraded its own equipment library: Intelligent Body Pro and enterprise-level MCP services .Zhu Guangxiang, general manager of Baidu Intelligent Cloud Application Platform Department, compared the Intelligent Agent Pro to the strongest librarian, senior researcher and computer geek . It can not only autonomously analyze problems, reason, search for information, summarize and reflect until a satisfactory answer is given, but also solve all-round research tasks such as data analysis, chart drawing and report writing, and even simulate people playing computers or operating business systems on the enterprise side, truly realizing full-stack autonomy and intelligence.Intelligent Body Pro has been implemented in the sewage treatment project information aggregation platform of Wastewater Treasure. The Agent customized for it needs to not only retrieve scattered and regional small and medium-sized projects, but also accurately connect with engineering units and bidding information.Based on a series of tools in the Qianfan big model platform, enterprises can input instructions to associate knowledge bases and then quickly generate customized agents. In actual operation, this agent can search for historical bidding data in the internal knowledge base of the enterprise, and then conduct multiple rounds of searches on millions of projects, and then generate a complete report based on public information on the Internet, which also contains traceable links.As one of the most popular protocols nowadays, MCP has realized a unified tool interface and lowered the development threshold. Qianfan Big Model Platform has also become one of the first platforms in China to fully embrace the MCP protocol. This time, it launched enterprise-level MCP services and connected to more than 1,000 MCP servers . Developers can independently develop MCP servers based on the tool chain provided by Qianfan Big Model Platform and publish them to Qianfan MCP Plaza with one click.This low-threshold, high-efficiency development model is fully activating the Agent ecosystem.Conclusion: One-stop empowerment breaks the deadlockActivate the application value of large models
Baidu Smart Cloud Qianfan Big Model Platform is a one-stop empowerment platform that is becoming the foundation for the current explosion of big model value, and is fully equipped to meet the needs of enterprise intelligent transformation. To date, the Qianfan platform has covered more than 400,000 customers, with a penetration rate of more than 60% in central enterprises, more than 45,000 fine-tuned models, and more than 1 million developed applications. Massive industry implementation practices continue to feed back to the platform iteration, forming a positive cycle of "application practice, technology optimization, and capability upgrade."At the same time, data from intelligent hyperparameters show that in the first quarter of this year, Baidu Smart Cloud achieved the first place in both the number of large model winning projects and the winning amount . These all prove that the Qianfan large model platform, which has everything from model calling, development to application development, and combines a full set of "resources + tools + scenarios" services, has made a breakthrough in reducing enterprise development costs and simplifying complex processes.The application of big models is surging. We have foreseen that with the prosperity of MCP ecology and super agents, the technical dividends of big models will truly be transformed into the actual value of enterprise cost reduction, efficiency improvement, and innovative growth. This revolution has also spawned new opportunities. Only by embracing change and actively transforming can we occupy a place in the business landscape reshaped by AI.
