2025年"AI Agent 之年"正逐渐走向现实。 3月初 Manus 的出现，真正让用户感受到了Agent产品与对话类大模型的本质区别：用户看到 AI 能够自主操作电脑、使用浏览器、编写代码。Manus 基本定义了 Agent 类产品的基本功能和产品 形态。 然而由于早期服务器资源的限制，Manus采用的邀请码注册形式，没有快速积攒很好的用户口碑， 最终没能像DeepSeek-R1那样将 AI Agent 的概念植入用户的日常生活。 但所幸字节跳动没有让用户等待太久。4 月 18 日，字节旗下首款 toC Agent 应用"扣子空间"正式 上线内测。 本文约 4300 字，推荐阅读时长 24 分钟，欢迎关注新财富公众号。 1 扣子空间实测：不俗的性能，完美的交付能力 字节对于扣子空间的定义，是精通各项技能的"通用实习生"与各行业的"领域专家"，具备无限拓展 能力的 AI Agent。 扣子开发平台（Coze）是字节在24年1月上线的低代码AI开发平台，由字节AI Seed 部门的 Stone 团队开发负责。扣子面向个人和企业用户，具有零代码开发、开放生态、任务自治等核心特点，致 力于让所有人都能快速、低门 ...

Core Insights - The article discusses the emergence of AI Agents, particularly focusing on ByteDance's "扣子空间" (Coze Space) as a significant development in the AI landscape, showcasing its capabilities and potential impact on user interaction with AI technology [2][53]. Group 1: Product Overview - "扣子空间" is defined as a versatile AI Agent that acts as a "general intern" proficient in various skills, aimed at both individual and enterprise users, emphasizing low-code development and task autonomy [4][29]. - The product has two development modes: Exploration Mode and Planning Mode, allowing users to customize AI-generated pathways or receive guided assistance [5][29]. - Initial tests of "扣子空间" demonstrated strong performance in generating travel plans and developing simple games, indicating its capability to handle medium-difficulty tasks effectively [3][16][19]. Group 2: Technical Features - "扣子空间" integrates various plugins, including Gaode Map and MySQL, enabling users to customize their experience and access data from ByteDance's ecosystem [41][44]. - The introduction of the Model Context Protocol (MCP) allows "扣子空间" to interact with different databases seamlessly, enhancing its functionality and user experience [44][49]. - The article highlights the importance of user input quality in maximizing the effectiveness of AI Agents, suggesting that simplifying the process of formulating requests could be a competitive advantage [29][48]. Group 3: Competitive Landscape - ByteDance is not the only player in the AI Agent space; other major companies like Alibaba are also developing similar products and integrating MCP into their offerings [50][53]. - The article suggests that the competition among AI Agents will increasingly focus on the completeness of tools and the performance of underlying models, rather than just the models themselves [49][52]. - The release of "扣子空间" is seen as a pivotal moment in educating users about AI capabilities and setting a standard for future AI Agent products [53].