Update/ As the competition for entry points in the AI era begins, the ecosystem is gradually becoming clearer. Having understood the four layers of LLM/Agent/MCP/Skill, I believe the next step will be entry points and applications very similar to the internet. 1/ Today, it seems OpenClaw was inevitable. The next step, such agents, may take different forms, from browsers and chat to consoles; a diverse range of approaches is unlikely, as paradigms are hidden within. 2/ Huang Renxun's computing power explosion and the endless capabilities of LLM mean that capability enhancement is moving from leaps to smoother progression, supporting scripting giants like Xiaolongxia (King of Scripts). Usability is paramount, and models can be varied. It's time to focus on user experience. Different types of platforms will gradually emerge, and based on this product hierarchy, LLM will offer computing power options, becoming increasingly cheaper. Even more powerful large-scale models will enter the market. 3/MCP has standard protocol specifications similar to traditional software service API interfaces, but its function is relatively implicit, making it difficult for developers to understand deeply. Skills are clearly defined as specific capability encapsulations, so the problem isn't a lack of ability, but rather a lack of imagination. The "Skill" designation is clear, comparable to a mobile desktop app, but open-source and flexible, it's a programmable plugin library encompassing both ideas and capabilities. 4/ Who made the most significant contribution? Of the four terms, MCP and Skill were coined by Athropic, a later entrant, clearly demonstrating their understanding of application. Whether based on capabilities like Videocode or advocating for a diverse range of AI skills, they focused on how to use them and who will use them. 5/ I remember watching a Stanford University video interview with Sam a few years ago, where he was asked how to implement these applications. Sam replied, "I don't know. It depends on the collective wisdom." Less than a year later, Athropic, with Cluade/ClaudeCode, launched a counterattack, taking over both OpenAI and Cursor. This is reminiscent of Microsoft's late-mover advantage, quickly acquiring user-friendly GUI Windows systems and the Word/Excel suite, and securing IBM OS/2, Apple Macintosh, and the leading office suite, Lotus.