Zhongtai: Manus may mark the beginning of the commercialization of universal Agent.

Zhongtai released a research report stating that on March 6, the Monica team launched the world's first universal Agent product, Manus. Manus is not a simple question-and-answer AI; its core advantage lies in its ability to autonomously plan and execute complex tasks. Manus has a multi-Agent architecture with three core modules: planning agent, executing agent, and verification agent. The target market for Agent is not just the traditional software market, but also the trillion-dollar service market. By transforming services into the software model (Service-as-a-Software), applications represented by Agent will create value far beyond traditional software. Universal Agent products like Manus will also begin to commercialize. Zhongtai's main points are as follows: Event: On March 6, 2025, the Monica team released the world's first universal Agent product, Manus. In the GAIA Benchmark, Manus performed at SOTA levels: Level 1 86.5% (OpenAI Deep Research at 74.3%), Level 2 70.1% (OpenAI Deep Research at 69.1%), Level 3 57.7% (OpenAI Deep Research at 47.6%). Multi-Agent architecture and integrated toolsets support complex task execution capability Manus is not a simple question-and-answer AI; its core advantage lies in its ability to autonomously plan and execute complex tasks. Manus has a multi-Agent architecture with three core modules: planning agent, executing agent, and verification agent. The planning agent breaks down the user's vague requirements into a series of executable subtasks and formulates the corresponding execution plan. The executing agent calls various external tools (such as code editors, browsers, specific applications, etc.) to complete the subtasks according to the plan. Finally, the verification agent checks the execution results to ensure that the final output meets the user's expectations. This decomposition mechanism enables Manus to truly have the capability to execute complex tasks. Strong intention understanding and interaction capabilities, feedback mechanism improves usability 1) Leveraging LLM reasoning to improve the user's intention understanding capability, even in the face of vague or incomplete instructions, Manus can understand and reason through context, automatically supplement missing information, and formulate reasonable execution plans. 2) The feedback mechanism is improved, providing real-time progress and results of task execution to the user and promptly handling any errors that may occur. This smooth interactive experience significantly reduces the user's learning curve, and there is also more potential for applications in low-code development, workflow automation, etc. As the Manus team believes in the philosophy of "less structure, more intelligence," when data is of high quality, models are strong, architecture is flexible, and engineering is solid, the concepts of Computer Use, Deep Research, Coding Agent, etc., will naturally emerge as capabilities. Manus may herald the dawn of the commercialization era for universal Agent On the enterprise (B) side, Agent can be considered as a decision-making RPA or an LLM capable of interacting with the external environment, greatly simplifying user workflows to reduce costs and increase efficiency in the short term. Applications in coding, marketing, and HR recruitment have already demonstrated high value. In the long run, Agents can create value directly in business operations or empower decision-making. Agents are aiming at and replacing not just the traditional software market but also the trillion-dollar service market. By transforming services into the software model (Service-as-a-Software), the application model represented by Agent will create value far surpassing traditional software. Perhaps from this moment, universal Agent products like Manus will also begin to commercialize. Key targets 1) Computing power: Recommended focus on Inspur Electronic Information Industry (000977.SZ), Unisplendour Corporation (000938.SZ), Dawning Information Industry (603019.SH), Digital China Group (000034.SZ), etc. 2) Model end: Recommended focus on Iflytek Co., Ltd. (002230.SZ), SENSETIME-W (00020), etc. 3) Application end: Continuous focus on AI Agent vendors such as Beijing Kingsoft Office Software, Inc. (688111.SH), Focus Technology (002315.SZ), KINGDEE INT'L (00268), MING YUAN CLOUD (00909), Digiwin Co., Ltd. (300378.SZ), Richinfo Technology (300634.SZ), Wondershare Technology Group (300624.SZ), Hithink RoyalFlush Information Network (300033.SZ), Hand Enterprise Solutions (300170.SZ), Shanghai Newtouch Software (688590.SH), Fujian Foxit Software Development Joint Stock (688095.SH), Intsig Inform. For artificial intelligence companies, investors can pay attention to SenseTime Group Ltd. (688615.SH), Beijing DeepGlint Technology (688207.SH), CloudWalk Technology (688327.SH), etc.; for intelligent driving companies, investors can consider HORIZONROBOT-W (09660), Weirong Hengrun (688326.SH), Huizhou Desay SV Automotive (002920.SZ), Thunder Software Technology (300496.SZ), ArcSoft Corporation (688088.SH), etc.; and for intelligent robotics companies, Siasun Robot&Automation such as Shanghai Suochen Information Technology (688507.SH), Nancal Technology (603859.SH), etc.Risk Warning The risk of AI technology not being implemented as expected, intensifying competition, and outdated information updates, etc.