Agentic AI在5000行以上代码失效的根本原因是上下文窗口限制。解决方案：用文档驱动开发作为长期记忆，支持多Agent协作和人机配合工作流。
Articles by grapeot
From Context Amnesia to Document-Driven Development: Overcoming the Project-Scale Trap in Agentic AI
Agentic AI fails above ~5000 lines due to context window limitations. Solution: document-driven development as long-term memory, enabling multi-agent collaboration and human-AI workflow.
Manus爆火的背后，Agentic AI产品如何构筑持久的竞争优势？
分析Manus如何打通调研和生成链路实现端到端体验，以及Agentic AI产品在工具、数据和智能三个维度的复利效应如何构筑持久护城河。
Behind Manus's Wild Popularity: How Agentic AI Builds Lasting Competitive Advantages
Exploring why Manus succeeded by bridging Deep Research and Cursor, and analyzing the compound effects in tools, data, and intelligence that create lasting moats for Agentic AI products.
被低估的Claude Code：AI Native软件开发的关键拼图
Claude Code是被低估的AI Native软件开发关键拼图，推动Library as a Service模式，帮助AI有效对接企业级代码库。