In-Depth: How Will AI Agents Transform Everyday Work and Life? (AI Translation)

【周刊提前读】深度：AI智能体应用如何改变日常工作生活？

After the global rise of DeepSeek, another Chinese company named Butterfly Effect has drawn global attention by introducing its AI Agent product, Manus, to the public. This move brings AI agents—a more autonomous form of AI capable of handling complex tasks—further into mainstream use, bridging the gap between human conception and execution. Manus distinguishes itself by autonomously screening job resumes and performing tasks like property research and stock analysis, delivering not just ideas but tangible results. Despite being released overseas, access to Manus is highly sought after in China, with invite codes being sold at high prices on secondary markets. Its founder, Ji Yichao, is a serial product developer with an entrepreneurial background but not a hardcore AI research pedigree. His prior ventures include the Mammoth browser and Peak Labs, acquired in 2022[para. 1][para. 2][para. 3].

Before Manus’s surge, the term “AI agent” was loosely used for applications integrating multiple functions to assist or replace humans in specific tasks, but was not yet a clearly defined product concept. The Lovable AI programmer, launched at the end of 2024, exemplifies the transition, with its broad programming abilities and rapid growth: earning $17 million in annualized recurring revenue within three months and attracting 30,000+ paying users with a small team. Investment capital has been flowing in, with Lovable raising $15 million in February 2025, led by prominent Europe-based VC Creandum and supported by industry leaders. The trend reflects a shift from single-content generation like language and images to more comprehensive, action-capable AI agents, as both tech giants and startups latch onto the agent paradigm amid active investment from both US and Chinese venture capital[para. 4][para. 5][para. 6].

The rapid proliferation of AI agents is driven by advances in underlying large models and the emergence of universal interaction protocols. Amazon-backed Anthropic and OpenAI have released increasingly sophisticated models—Claude 4, OpenAI’s o3 and o4-mini, and Google’s Gemini 2.5 Pro—all extending agent capabilities, such as long-running, multi-step reasoning and task execution. In China, DeepSeek upgraded its R1 model, ByteDance’s Volcano Engine released AI tools for agent development, and traditional internet giants Alibaba and Tencent are building their agent ecosystems[para. 7][para. 8].

AI agents are quickly moving from enterprise workflows (web development, code testing) to consumer tasks (multi-app coordination, complex task automation). Tech giants each define agents as software capable of planning, reasoning, and executing tasks to maintain user productivity, reflecting a global consensus that agents will increasingly resemble human collaborators in complex, long-duration tasks. Supporting vertical tools like UI designer MiaoDuo and creative platform LibLibAI’s Lovart indicate rapidly improving AI coding and creative abilities, reducing repetitive manual work and expanding final deliverable quality[para. 9][para. 10][para. 11][para. 12].

Despite rapid development, the usability and efficiency of AI agents remain works in progress. Challenges include low-quality code generation, time inefficiency, and the need for human oversight. Currently, Gartner divides agents into chatbots ("conversational"), current agents ("assistive/action-oriented"), and future self-learning agents capable of independent problem-solving in uncertain environments. Technological efforts like Google DeepMind’s AlphaEvolve aim for agents to “evolve” superior solutions autonomously[para. 13][para. 14][para. 15].

Agent ecosystems are forming around open protocols. Anthropic’s Model Context Protocol (MCP) standardizes agent connection to external data systems, while Google’s Agent2Agent Protocol (A2A), released in April 2025, facilitates agent-to-agent communication and collaboration, supported by major enterprise partners. MCP’s adoption is visible across both US and Chinese tech companies. Nevertheless, most ecosystems remain partially isolated, with key servers and applications missing from universal integration, reflecting underlying business competition[para. 16][para. 17][para. 18].

The large model price war is fierce: OpenAI cut prices by 80% for model o3-pro, ByteDance’s Doubao model follows suit, narrowing profit windows to months instead of years. Tech giants are now restricting access to their core data and APIs, as seen with Microsoft Bing’s tightening around Chinese AI companies, forcing domestic companies to develop alternatives and reduce dependencies on foreign data services[para. 19][para. 20][para. 21].

AI agent startups have become new investment darlings, especially for US venture capital and Chinese internet companies. Startups like Manus and LibLibAI have raised substantial rounds, often reaching valuations of hundreds of millions of dollars prior to product launches, if led by credible founders. However, competitive risks from internet giants, barriers in data access, and the need for viable business models make survival challenging for startups. The trend is for specialist, vertical agents rather than general-purpose ones, as big companies dominate major consumer entry points[para. 22][para. 23][para. 24][para. 25][para. 26][para. 27][para. 28][para. 29].