发现 AI 产品的关键不只是模型，而是人如何理解和控制系统。I learned that AI products depend not only on models, but on how people understand and control the system.
设计 AI Agent UI、状态面板和工作流界面。Designed AI agent interfaces, status panels, and workflow views.
Journey Map
从这张图里看见我把想法做出来A map of how ideas
become working projects
Codex Harness / OS BrainCodex Harness / OS Brain
研究个人化 Harness 系统。Explore a personal Harness system.
AI 工具需要记忆、上下文和反馈机制。AI tools need memory, context, and feedback mechanisms.
点击上方节点，不是看流水账，而是看我每一步怎么换问题、换工具、换表达方式。Use the nodes to see how the question, tools, and form of expression changed at each stage.
我的工作方式更像一张循环草图My working method is a continuous loop
先把问题说清楚，再把结构画出来，然后用原型和反馈继续逼近真实价值。Clarify the problem, draw the structure, then use prototypes and feedback to move closer to real value.
Decode
先把真实问题拆出来，避免一上来就做漂亮但没意义的 Demo。Separate the real problem before building a polished but meaningless demo.
Structure
把用户、流程、限制和失败场景理清楚，让想法有可执行边界。Clarify users, workflows, constraints, and failure cases so the idea has executable boundaries.
Visualize
把抽象系统变成界面、流程图和状态面板，方便讨论和判断。Translate abstract systems into interfaces, flows, and status panels that support discussion and judgment.
Prototype
用前端和 AI 工具做成可验证原型，而不是停在概念稿。Build a verifiable prototype with frontend and AI tools instead of stopping at a concept.
Reflect
复盘问题，把经验沉淀成 Prompt、工具链和下一次可复用的方法。Review failures and turn experience into prompts, toolchains, and methods that can be reused next time.