An engineering notebook on AI, systems, and everyday experiments.

探索AI如何处理非结构化语料，利用百万级上下文窗口让AI阅读三年微信群聊记录，并通过反思-提取-构建的循环协作方式，从中提炼出关于AI协作的深刻洞见与结构化知识。

Fed 3 years of messy WeChat history into GPT-4.1's 1M token window—it synthesized themes, generated insights, and wrote a cohesive book about AI collaboration.

四步实现GPT文生图的商业级精准可控：用视觉撬动AI思维、用遮罩迭代生成、让AI当合伙人、用文档管理打造AI资产库，生成可复用的统一视觉风格。

Four techniques to achieve commercial-grade flowcharts and visualizations with GPT's image generation: visual prompts, masking for iteration, AI brainstorming, and document management for reusable assets.

分析MCP解决LLM工具调用标准化痛点的技术优势和Anthropic的商业推动策略，以及开发者何时应该使用MCP、何时可以不用。

Examining why MCP gained popularity for standardizing LLM tool invocation, its technical advantages and limitations, and when developers should adopt it versus building without frameworks.

从性能SUV换到劳斯莱斯库里南后，发现人和车相互塑造影响，油门刹车调校、底盘滤震和路感都在潜移默化改变驾驶风格和心态。

How switching from a performance SUV to a Rolls-Royce Cullinan changed my driving style, revealing the mutual influence between users and products through throttle, chassis, and road feel.

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。

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.