我在医疗行业干了十几年，到今天，还写不出一行代码。
这话有点没面子，但它恰恰是我开这个号的原因。
过去这一两年，我没写一行代码，却靠着 AI，把自己干了十几年的活——医疗推广——一件件重做了一遍。不是"做几张图、改几段文案"那种用了一下 AI，是把过去只能靠堆人、靠个人手感的推广，做成了能跑、能复制、会自己迭代的系统：医疗短视频的全链路生产线；自己选题、自己制作、自己发布、自己复盘的内容起号闭环；一套既要讲得专业、又要能过平台合规的医生科普编导方法论；还有一个自动盯着市场动静的情报库。这些不是 PPT 上的构想，是每天都在跑的东西。
很多人第一反应是：不写代码，你怎么落地 AI？
我的答案正好反过来。AI 在医疗落地，门槛从来不是技术。
真正的门槛，是你懂不懂这行的规矩。
我做的这一端，是医疗营销里最难啃的一块骨头：处方药的线上推广。它不像面膜、不像消费品，可以直接对着大众投。它受严格的广告和合规约束，你真正要解决的是怎么合规地触达到该看到的人，又怎么把声量变成真实的处方和市场份额。这里面有一堆看不见的红线：什么能说、什么不能说，怎么触达才算合规，怎么在不越界的前提下把话说到位。这些东西不写在代码里，写在十几年趟过的雷里。
懂医疗的人，多半不知道怎么把 AI 用起来；懂 AI 的人，多半不懂医疗这套规矩和推广打法。我恰好卡在这条缝上。
说个具体的。我搭那套医生科普编导系统时，最难的从来不是调哪个模型、用哪个工具，是把两件看着对立的事同时做到：一句科普既要专业、不出错，又要能过平台那关、不被判成违规广告。一开始我靠人一条条盯，盯到眼睛花。后来我把这套判断拆成机器能执行的步骤：哪些表达安全、哪些一碰就限流、怎么改既保住专业又躲开红线，让它先自己过一遍。踩过的坑不少：有的稿子机器觉得没问题，真发出去还是被限流，我就回头把那条规则再抠细。改了几轮，它才真稳下来。这一步说实话，只有真在这行里挨过罚的人才做得出来。
我为什么现在要把这些明着写出来？
因为我看到一个说法，挺戳我的：在传统行业里学 AI，不能偷偷学、偷偷用。你学会了、但没人知道，约等于没学会。所以我决定明着做，把我怎么一步步做出来的、包括踩过的坑，都摊开写。
还有个判断我想说在前头：现在谈医疗 AI，钱和目光几乎全在两件事上：AI 做药物研发、AI 做辅助诊断。可医疗有一整侧几乎没人认真碰：医疗到底怎么合规地触达人、怎么跑通推广和商业这台引擎。研发和诊断当然重要，但它们周期长、离回报远；推广和商业化这一侧，恰恰最快能出成果、离钱最近，却被晾在一边，没人正经碰。处方药推广，又是这侧里最硬、最少人碰的地方。难，所以大多数人绕着走；也正因为难，谁先用 AI 把它做通，谁就拿到了别人拿不到的东西。
这个号往后就聊这件事：医疗 AI 该往哪走，尤其是被忽略的推广和商业化这一侧；以及一个不写代码的医疗人，怎么把 AI 真正落地成在跑的系统。不讲玄的，只讲我看到的趋势，和我自己亲手验证过的东西。
如果你也是这行里那个"懂业务、却被技术挡在门外"的人，我想早点告诉你：门槛没你想的那么高。你手里那套别人没有的行业理解，才是 AI 时代最值钱的东西。
第一篇就到这。往后见。
I've worked in healthcare for over a decade, and to this day I still can't write a single line of code.
That's a bit embarrassing to admit, but it's the whole reason I started this account.
Over the past year or two, without writing any code, I used AI to redo the work I'd spent more than a decade on, healthcare promotion, one piece at a time. I don't mean the "make a few images, tweak some copy" kind of dabbling with AI. I mean taking promotion that used to run on manpower and personal instinct and turning it into systems that actually run, can be copied, and improve themselves: a full production pipeline for medical short videos; a content-launch loop that picks its own topics, produces the videos, publishes them, and reviews its own results; a method for directing doctors' health-science videos that has to be both medically sound and able to clear platform compliance; and a market-intelligence base that keeps an eye on things on its own. None of this is a concept on a slide. It's all running, every day.
A lot of people's first reaction is: if you don't write code, how do you get AI to actually work?
My answer is the exact opposite. When it comes to getting AI to work in healthcare, the barrier was never the technology.
The real barrier is whether you understand the rules of this industry.
The corner I work in is the hardest nut to crack in healthcare marketing: promoting prescription drugs online. It's not a face mask or a consumer product you can advertise straight to the public. It's under strict advertising and compliance rules, and what you really have to solve is how to reach the people who should see it without breaking those rules, and how to turn attention into real prescriptions and market share. There's a pile of invisible red lines in here: what you can say and what you can't, what kind of outreach counts as compliant, how to get the message across without stepping over the line. None of this is written in code. It's written in the landmines I've stepped on over a dozen years.
People who know healthcare usually have no idea how to put AI to work. People who know AI usually don't understand healthcare's rules or how promotion works here. I happen to sit right in that gap.
Here's a concrete example. When I built the system for directing doctors' health-science videos, the hardest part was never which model to tune or which tool to pick. It was making two things that look like opposites hold true at once: a piece of health-science content has to be accurate and genuinely expert, and it also has to clear the platform's review without getting flagged as a non-compliant ad. At first I checked every line by hand, until my eyes went blurry. Then I broke that judgment down into steps a machine could follow: which phrasings are safe, which ones get you throttled the moment you use them, and how to rewrite so you keep the substance and stay clear of the red line. I let the machine take the first pass. Plenty went wrong along the way. Some drafts it cleared still got throttled once they went live, so I'd go back and tighten that rule. It took a few rounds before it really settled down. And honestly, this is the kind of thing only someone who's actually been penalized in this business can build.
So why am I putting all this out in the open now?
Because I came across a line that stuck with me: when you're learning AI inside a traditional industry, you can't do it in secret. If you've figured it out but nobody knows, that's about the same as not having figured it out at all. So I decided to work in the open, and to lay out how I did it step by step, including everything I got wrong.
There's one more point I want to make up front. When people talk about healthcare AI right now, almost all the money and attention go to two things: AI for drug discovery and AI-assisted diagnosis. But there's a whole side of healthcare that hardly anyone is seriously working on: how you actually reach people within the rules, and how you get the whole promotion-and-business engine running. Drug discovery and diagnosis matter, no question. But they run on long timelines and sit a long way from any payoff. The promotion side is the one that can show results fastest and is closest to the money, and it's the one everybody's leaving alone. Prescription-drug promotion is the hardest part of it, and the part almost nobody touches. It's hard, so most people walk around it. And because it's hard, whoever gets AI working there first ends up with something nobody else has.
From here on, that's what this account is about: where healthcare AI should be heading, especially the promotion and commercial side everyone overlooks, and how someone in healthcare who doesn't write code can turn AI into systems that actually run. No hand-waving. Just the trends I'm seeing and the things I've tested myself.
If you're the person in this business who knows the work but keeps getting shut out by the technology, let me tell you sooner rather than later: the barrier isn't as high as you think. That industry understanding you're holding, the kind nobody else has, is the most valuable thing there is in the AI era.
That's it for the first piece. See you in the next one.