把 AI 想清楚，并且公开地想。
I’m MJ, an AI product manager who builds. I write about context engineering, LLM behavior, and what separates successful AI adoption from expensive experiments.我是 MJ，一个会动手做东西的 AI 产品经理。我写上下文工程、大模型的行为，以及企业里 AI 落地成功和烧钱试错之间的差别。
Every run since 2021, synced from Strava.2021 年以来的每一次跑步，从 Strava 同步。
A paper-trading portfolio, priced every weekday.一个模拟盘组合，每个交易日更新价格。
I think deeply about how AI products actually work: why LLMs hallucinate, how context engineering shapes behavior, and what separates successful AI adoption from expensive experiments.我关心 AI 产品到底是怎么运转的：大模型为什么会幻觉，上下文工程怎样塑造模型的行为，以及 AI 落地成功和烧钱试错之间差在哪里。
This site is where I work through ideas in public, grounded in real experience shipping AI products and helping teams navigate the messy reality of enterprise AI adoption.这里是我公开想问题的地方。想法来自真实地交付 AI 产品，也来自陪团队走过企业 AI 落地的种种混乱。