Zhongtai: Manus' universal intelligent body is initially formed, opening a new era for AIAgents.

Zhongtai released a research report stating that on March 6, 2025, the first general-purpose AI intelligent entity Manus was born, which can break down complex tasks into executable steps and call tools flexibly in a virtual environment, ultimately delivering complete results. This initially demonstrates the independent working ability of the Agent in form, marking the transition of AI from Reasoner to Agent and the entry of Agent's ability into the "Take Actions" stage. The entry of Agent competition is king, and manufacturers with more user traffic are expected to achieve a positive cycle of "traffic-data-usage experience". With the upgrading of open-source model capabilities, the technology gap between large and small companies can be filled, and the engineering capabilities of AI products may widen the gap in product user experience. Zhongtai's main points are as follows: Agent = digital worker, achieving autonomous decision-making and task execution AI Agent = LLM (planning + memory + tools + action), can perceive the environment, make autonomous decisions, and carry out actions. It has the ability to gradually achieve goals through independent thinking and tool calling, not limited to another form of applications, but a direct comparison to human labor. Currently, Task Agent is the mainstream, and Multi-Agent collaboration applications experience improvement. The Agent market potential is sufficient, helping industries reduce costs and reshape value. The strong willingness in professional services, finance, healthcare, and retail scenarios may lead to the opening of a trillion-dollar Chinese AI market. Singularity is approaching: Manus, a universal intelligent entity, is in the initial stage, opening a new era of AIAgent The development focus of large models has shifted towards "post-training, reasoning side, autonomous intelligent entity" from "convolutional base, convolutional model" to "convolutional reasoning, convolutional AIAgent, convolutional application landing". On March 6, 2025, the first general-purpose AI intelligent entity Manus was born, which can break down complex tasks into executable steps and call tools flexibly in a virtual environment, ultimately delivering complete results. In form, it initially demonstrates the independent working ability of the Agent, marking the transition of AI from Reasoner to Agent and the entry of Agent's ability into the "Take Actions" stage. Industry Outlook: Focus on entrances, AI applications + intelligent hardware usher in industrial revolution AI applications: traffic is king, product engineering capabilities are the focus for future development. The entry of Agent competition is king, and manufacturers with more user traffic are expected to achieve a positive cycle of "traffic-data-usage experience". With the upgrading of open-source model capabilities, the technology gap between large and small companies can be filled, and the engineering capabilities of AI products may widen the gap in product user experience. 1) General vs. specific: General-purpose Agents require higher generalization capability, and platform-based companies have higher research and development endowment and have a larger user base, a wide product matrix, sufficient data accumulation, and abundant cash flow. They have a greater advantage in the competition of general-purpose Agents. Specific Agents need to strengthen their understanding of specific fields, and relevant specific industry manufacturers may differentiate themselves through customer resources, business understanding, and specific data accumulation. 2) ToC vs ToB: C-end Agents focus on product user experience, ultimately competing in traffic and payment rates; B-end Agents focus on enterprise tasks, aligning with the strong cost reduction and increased efficiency intentions of enterprises. Moreover, the specific business scenarios, business logic, and industry data of enterprises are more suitable for the use of Agents. Intelligent hardware: The endpoint realization of Agents is the last mile of intelligent entities, and various types of intelligent hardware are flourishing. As AI Agent technology iterates and industrializes, intelligent hardware products may face the opportunity of industrial upgrading, becoming the "entry point" for general or specific areas. Investment advice: Application side: Cost revolution at the endpoint + autonomous intelligence, Agent applications are expected to explode. The open source wave of DeepSeek has sparked a cost revolution, model capabilities upgrade accelerating the landing of applications, and the comprehensive outbreak of intelligent tool products. Recommended investments include: Hithink RoyalFlush Information Network (300033.SZ), Beijing Kingsoft Office Software, Inc.(688111.SH), Intsig Information (688615.SH), Hangzhou EZVIZ Network (688475.SH), etc. Computing power side: The surge in application demand corresponds to the growth in computational consumption. Public cloud + private cloud + all-in-one machine as the three mainstream deployment forms drive the benefits of domestic core computing power. 1) Public cloud: Cloud computing giants increase CAPEX expenditure, driving computational power enhancement and effective computational power adjustment. Recommended investments include: Alibaba (09988), Tencent (00700), etc. 2) Private cloud: Private deployment is a national necessity, with increased procurement intentions. Recommended investments include: Hygon Information Technology (688041.SH), Dawning Information Industry (603019.SH), Shenzhen Sed Industry (000032.SZ), etc. 3) All-in-one machine: Lightweight deployment for party and government scenes and financial industries, with high demand. Recommended investments include: Digital China Group (000034.SZ), iSoftStone Information Technology (301236.SZ), Fiberhome Telecommunication Technologies (600498.SH), Talkweb Information System (002261.SZ), Zhejiang Dahua Technology (002236.SZ), etc. Risk warning: AI product engineering capabilities fail to meet expectations; commercial progress falls short of expectations.The progress of model technology iteration falls behind expectations; there is a risk of information lag or outdated information in the public data used in the research report.Pourriez-vous m'aider avec ces documents s'il vous plat ?