Original video in Chinese.
Key Takeaway
- You have to get hands-on with Claude Code yourself: even if you get banned frequently, you still need to use it. It represents the real Agentic AI era and is completely different from ordinary chatbots. Only firsthand experience can build the right understanding; you can’t just watch videos or use the free version and draw conclusions.
- Completely solve the infrastructure problem: the most advanced AI is overseas (China is about a year behind), so you must prioritize infrastructure like payments and access first, such as getting a Visa/MasterCard, a Capital One credit card, and so on. If you don’t solve this, no matter how good the tools are, you still won’t be able to use them.
- Markdown-ify your materials + Skill-ify your capabilities: convert all your personal materials into Markdown so AI can understand them; at the same time, package your own experience and workflows into Skills/SOPs, so humans and AI can become a “cognitive community.” This is the key shift from a consumer mindset to a builder mindset.
If I were starting over and learning AI again, these are the three things I would definitely do. I’ve actually said all of this repeatedly in the Knowledge Planet before. In this video, I’m laying it out completely.
First, use Claude Code. Even if you get banned in less than two days, you still have to use it!
Today, we have already entered the Agentic AI stage. In this stage, the most advanced and most representative tool is Claude Code. It’s completely different from the chatbots before—it’s worlds apart!
Chatbots output text and images, which is easy to understand. But tools like Claude Code create more things through code and control more things through commands. You simply can’t imagine that just from other people’s descriptions or by watching a few videos.
So if I were starting over today, the first thing I would do is personally get hands-on with Claude Code.
Only by using it yourself do you get a real feel for it.
AI is developing at breakneck speed. That sounds cliché, but I really can’t think of a more accurate phrase than “breakneck speed” to describe it. In this situation, you absolutely cannot sit in your own little well and think you know everything—for example, thinking AI is nothing special.
You have to stick your head out and take a look yourself! Only by seeing it with your own eyes do you know how vast the sky really is!
That’s why I said earlier that even if you get banned, you still have to use it. Whether or not you keep using Claude Code afterward, in the beginning you absolutely have to experience it and get your hands on it.
Let me say a bit more. I previously made a video introducing Google Stitch. At the time, I judged that the people most likely to be put out of work next would be designers. A lot of people wouldn’t accept it and felt I was just spreading anxiety. Then, a couple of days ago, Claude Design came out.
You think design is pure imagination, or so-called intuition? Just wait and see: starting with the most standardized work, AI will step by step eat into designers’ share.
I’m going off on a tangent. Let’s move on to the next point.
Second, completely solve the infrastructure problem.
The most advanced AI is overseas. China is probably about a year behind.
That’s just a fact. It has nothing to do with patriotism or whether you support domestic AI.
For well-known reasons, to use overseas AI stably, you really need a set of infrastructure. This is also a long-running topic in our Knowledge Planet. I keep getting asked about it, and I keep emphasizing it.
So if I were starting over today, I would definitely spend time solving this first. Otherwise, the Agentic AI I just talked about won’t be usable either.
For example, the payment issue.
The simplest way is to get a domestic Visa or MasterCard credit card. Even though it’s a domestic card, it’s still quite useful. For example, I’ve been using a MasterCard from China Merchants Bank to subscribe to Gemini, and a Visa from China Construction Bank to subscribe to GitHub Copilot.
As for AI services that can’t be subscribed to with a domestic card, you can pay by using a U.S. iOS account plus gift cards.
Going one step further, you can also get a Capital One secured credit card. That’s a foreign card. Since I applied for an ITIN, getting all of this done went very smoothly.
To be honest, the thing I see most often is people being unable to use the most advanced AI because of infrastructure issues. I don’t know if it’s laziness or what, but they just refuse to fully solve it, and then they get stuck again and again. I really can’t understand it.
Third, Markdown-ify your materials, and Skill-ify your capabilities.
Everyone is talking about learning AI. But what does it mean to have truly learned it? There has to be some standard, right?
After I published a tutorial on learning AI, I also posted something like this in the Knowledge Planet about the standard:
I think that at this stage, there is only one benchmark: can the individual and AI become a “cognitive community”?
On the one hand, the individual clearly understands the boundaries of AI. These boundaries include the limits of AI capabilities and knowledge, the differences between different models and tools, and so on.
On the other hand, AI can dynamically anticipate based on its understanding of the individual.
To do that, you have to let AI understand you.
So if I were starting over today, I would definitely turn all the materials I’ve saved into Markdown format. Because this format is understandable to AI, and we humans can read it too.
As for Word, PDF, and PPT, I definitely wouldn’t bother with those anymore. They’re all too complicated, and AI can’t understand them. Since AI can’t understand them, these formats are bound to become more and more marginal.
In addition, I would Skill-ify all of my capabilities.
Honestly, that process is pretty difficult. Turning what’s in your head into a Skill, into a clear SOP, into text that AI can understand too—this is actually a process of self-improvement.
That’s also why, on New Year’s Day this year, I suggested in the Knowledge Planet that everyone create three Skills. Not only can that empower your own AI system, it can also thoroughly organize your workflow.
These three suggestions are all summaries of my experience over the past two or three years, and they’re things I’ve repeatedly emphasized in the Knowledge Planet. As for how to do it specifically, go search, go ask AI—don’t ask me. If you can’t figure it out, then don’t even think about learning AI.
OK, that’s all for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you in the next episode!