Recently, the Beijing Municipal Development and Reform Commission and four other departments issued the "Several Measures for Accelerating the Leading Development of Intelligent Agents in Beijing" (referred to as the "Several Measures"). This is the first provincial-level special policy for intelligent agents in China.

The "Several Measures" consist of 4 aspects and 10 articles, aiming to achieve higher development goals for intelligent agents in the next stage, focusing on technology, application, industry, ecology and other four aspects, and further forward-looking layout and leading development.

One is technological leadership, building a solid foundation for the development of intelligent agents. On the one hand, we will continue to promote the development of basic models. Recently, Professor Tang Jie, the founder of Zhipu, announced the "Touch High Plan", which is also a consensus in Beijing. He hopes that elites from all parties can jointly set ambitious goals for continuous "touch high", support the innovation of model architecture and training paradigms in online learning, continuous learning, autonomous evolution, and other aspects, and continuously break through the limit of intelligence. On the other hand, we will strengthen the research and development of common technologies in the control layer of intelligent agents. Support the implementation of control layer engineering, focusing on optimizing contextual engineering, task persistence, multi-agent collaboration, and system scalability to solidify the common foundation of intelligent agents. Accelerate the research and development of universal intelligent agent technology across models and chips, improve the length of complex task links and the stability of long-range task execution, build autonomous and controllable interconnection protocols, development frameworks, and basic toolchains, develop standard specifications, support the construction of intelligent agent infrastructure such as "Lingji", enhance the ability of intelligent agent development and deployment and cross framework interoperability, and ensure safety and controllability.

The second is to lead by application and accelerate the formation of a closed loop of value creation. On the one hand, it accelerates the implementation of native applications for intelligent agents. Support the development of AI native super software and artificial intelligence operating systems, develop scientific intelligent assistants and platforms with information association integration and knowledge system construction capabilities, create AI native autonomous laboratories, develop data-driven, autonomous decision-making, and human-machine collaboration industry solutions, promote the deployment of cutting-edge engineer (FDE) and other models to expand the depth of industry application landing, and form more benchmark demonstration scenarios in the fields of science, medicine, manufacturing, etc. On the other hand, it promotes the integration and development of intelligent terminals and intelligent agents. Enhance the environmental perception, autonomous decision-making, task execution, and lightweight deployment capabilities of intelligent agents, embed intelligent agent capabilities into various terminal hardware, develop new generation intelligent terminal products such as smartphones and smart glasses, promote the integrated development of chip and model cloud, encourage deep collaboration between chip companies, model manufacturers, and terminal enterprises, and continuously improve the ability to define future product forms in response to market demand.

The third is industry leadership, forward-looking cultivation of new forms of intelligent economy. On the one hand, it supports new entrepreneurial models represented by OPC and cultivates new organizational models for enterprises in the intelligent era. Accurately aligning with the characteristics of OPC, a "new species" of entrepreneurship in the intelligent era, constructing a hierarchical and classified professional service system, and upgrading public services around elastic computing power supply, professional incubation, etc. In response to the demand for admission and operation of OPC, we actively carry out pilot explorations to make OPC more convenient to implement and thrive. On the other hand, we encourage the development of Token economy and explore new paradigms for economic growth in the era of intelligence. Focusing closely on the inherent needs of the Token economy, we support the development of dedicated inference chips, carry out key technological breakthroughs such as inference architecture, accelerate the optimization of application layer efficiency such as inference cache reuse and intelligent task routing, and improve Token efficiency throughout the entire chain. Cultivate new business models such as Token as a Service (TaaS), Agent as a Service (AaaS), and Result as a Service (RaaS), and promote more Token products and services to be included in the scope of small and medium-sized enterprise service voucher distribution products. At the same time, encourage the transformation of token billing from quantity to quality, improve economic quality and efficiency, and perfect the token economic monitoring and evaluation method system.

The fourth is ecological guidance, creating a world-class innovation ecosystem. On the one hand, coordinate security governance and open source development. Enhance the governance capability of intelligent agents, further standardize the development of the intelligent agent industry, avoid malicious abuse of technology, develop security models to improve the ability to "use models to govern", accelerate the improvement of common service capabilities such as vulnerability scanning, security detection, attack and defense confrontation, and security authentication of intelligent agents. Build an open source community with global influence, support innovative entities to open source independent interconnection protocols, intelligent agent development frameworks, intelligent terminal operating systems, hardware public versions and other core codes, and encourage developers to jointly build and share the open source ecosystem. Building a high-level China SCO artificial intelligence application cooperation center, strengthening the public service system for going global, supporting more enterprises to go global, participate in international competition, and explore larger markets. On the other hand, we will strengthen the guarantee of key elements such as talent, computing power, and capital. In terms of talent, we will not only increase the cultivation of native and high-end talents to solve the problem of talent shortage, but also strengthen general skills training to enable more workers to make good use of artificial intelligence and intelligent agent tools. In terms of computing power, we will lay out the entire chain from power, chips, communication, interconnection, Infra, etc., strengthen the collaboration between computing and electronics, software and hardware, optimize the ultimate engineering, fully meet the high-frequency concurrency and low latency requirements of intelligent agents, and enhance the potential exploration of existing computing power. Support the construction of "Token Factories", increase support for computing power coupons, and encourage districts with conditions to explore the issuance of Token coupons, intelligent agent service coupons, etc., to provide more enterprises with inclusive and high-quality computing power services. At the same time, support financial institutions to develop specialized financial products that support the implementation of intelligent agents.