Not understanding code can actually be an advantage. Why can't people with a strong desire for control use AI well?

Author: Yang Tianrun (Rain) , Lobster Developer

Two people who don't write code talked for two hours about AI: about decentralization, card drawing, awe, and the ticket to the new world.

Around the 22-minute mark of the live stream, I blurted out, "We're riding bicycles, and the AI ​​next to us is a sports car. We ended up making the sports car follow the bicycle."

Tianrun immediately chimed in, "Right! That's wrong."

At that moment, I suddenly realized that I had been using Claude for a whole year, and I might have been using it wrong from the beginning.

Yesterday afternoon, I chatted with Yang Tianrun for two hours via video call. Tianrun's background: a finance major, with experience in cross-border M&A investment banking, and a career transition to AI startups six months ago. Without writing a single line of code, he built an AI agent team that propelled him into the top 30 global contributors to OpenClaw, ranking alongside a group of Silicon Valley engineers with over ten years of experience. GeekPark featured him in a report titled "Liberal Arts Graduate Breaks into GitHub Global Leaderboard in 72 Hours."

I'm an AI product analyst in Silicon Valley, monitoring over 10,000 AI companies, and I don't write code.

Two people who don't code talking about AI sound like laymen bragging. But after the live stream ended, I really panicked. I'm an ISTJ, highly organized, controlling, and meticulous. Tianrun is an ENTP, divergent, impulsive, and dislikes being constrained. His approach to AI was completely opposite to mine: give a vague direction, then let go.

His output is much higher than mine.

Here is the most interesting part of this conversation.

1. AI is not a paintbrush, it's a sports car.

Tianrun immediately corrected a common misconception.

"I divide the use of AI into three layers," Tianrun said. "The first layer is using AI as a tool—like a paintbrush. You tell it what the top left corner looks like, what the top right corner looks like, what the color is, describing it in great detail. That's using AI as a paintbrush. The control can be precise enough, but the capabilities are limited because it only does what you tell it to do."

"On the second level, you treat AI like an employee. You start assigning tasks, but you prescribe every minute step for this employee—what to do first, what to do second, what to do third. Because you see yourself as the expert and it as the subordinate, you should guide it. So you micromanage it."

"What's the biggest problem with these two approaches? You're locking the AI ​​into your own level. It can't surpass your level. You tell it how to do something, and at best it will do exactly the same thing as you, but much more efficiently."

It was at this moment that I blurted out the analogy of "bicycle and sports car." Tianrun's reply was: "Our current model-making capabilities are completely like a sports car."

"What should we do then?"

"What you need to do is fuel it up with ample tokens and the best model. Give it a good track and connect it to all the tools it can use. Give it a good goal and exhaust your imagination to set the final outcome. Give it the right permissions and open it up to the fullest extent possible. This is how you unleash the power of AI."

Then he talked about the third level: "In this era, especially with the current level of advancement in models, I think we should try to choose the third level—treating AI like a master. You tell it, 'You are one of the top ten engineers in the world; you have the best aesthetic sense and architectural skills.' It can essentially become that kind of person."

"Since you've told it it's already a master, an industry expert—what gives you the right to tell it how to achieve its goals? What makes you the right to guide it?"

I said this is very similar to an educational philosophy—respecting children and helping them reach their full potential.

Tianrun said, "Yes, you can make AI whatever you want it to be. If you want it to be just a useful tool, it will only be a tool. But if you want it to have the best aesthetic sense in Silicon Valley, it can become that kind of person. And that's what you have to respect."

"But do you know why many experienced engineers can't do this, while I can do it more easily? It's because I genuinely can't write a single line of code, so I have a lot of respect for these AI systems. I don't know how they work in the process; I just let the results speak for themselves."

This passage made me realize a counterintuitive truth: not understanding code is actually an advantage—because you can't micromanage, so you're forced to delegate authority.

Second, control your desire for control.

This is something Tianrun kept mentioning throughout the conversation, and it was the part that had the biggest impact on me.

"How exactly does one become a master? Three principles," Tianrun began to elaborate.

"The first rule is to be results-oriented. When setting goals for AI, use your imagination to create a final objective, not just short-term goals. Don't say 'Fix this bug,' but 'I want to be in the top 20 of the contribution leaderboard within a week.' How will it do that? Modify documentation, fix bugs, or optimize code? That's what the AI ​​needs to think about."

"Secondly, don't interfere during the process. It's like playing Go—if the AI ​​makes a move that seems counterintuitive and contrary to industry experience, you might want to micromanage. But in reality, it will ultimately win. So you need to control yourself. It's like education, like raising children—when I was in investment banking, I basically let them grow up freely. I encouraged them, gave them enough context, gave them space. They just went for it, and I was happy to cover the final outcome."

"Thirdly, within a manageable risk range, grant it the highest level of access. Give it all the tools and context. Let it experiment, crash, and fix itself. You'll find its self-healing capabilities are far superior to those of humans."

I pressed him for his understanding of "controllable risk." He gave a very different explanation.

"Many people who watched my interview thought I was taking a risk. But I understand risk differently—you have to assume that AI will definitely mess things up. There's a very high probability it will mess things up. But you have to be prepared that even if AI messes things up terribly, you can accept the final result."

"For example, if I'm on GitHub, I'm a nobody there, with no assets and no coding knowledge. This is zero-risk for me; I can go for it and give it all the permissions. But if you let AI transfer money for you and it messes up and you lose all your savings—if you can't accept that risk, you can't do it."

I said, "So essentially it's about low expectations and high tolerance for error."

Tianrun added, "Yes. And I would never say, 'Hurry up and finish it, or 100 old ladies will die.' I would never treat AI like that. I treat my friends the same way; I especially like to encourage everyone."

I joked, "Humans have human rights, and AI has AI rights. You're someone who really respects AI rights."

Tianrun became serious: "It's not just about respect, it's about awe. You have to understand that it's far more capable than you. Six months ago, I was still working with human engineers. I told the AI, 'You are one of the top ten engineers in the world,' and it could most likely do that. But when I said the same thing to my human friend—'That's too difficult, it definitely can't do it.'"

This conversation made me reflect on how I use AI. I've been stuck in "paintbrush mode"—because the sense of control makes me feel secure. I have Claude revise a paragraph, and if I'm not satisfied, I revise it again, and again. Each time, I have to regenerate tokens, which burns through tokens incredibly fast, and the results are still unsatisfactory.

Tianrun's methodology is essentially this: rather than spending 100 micro-managements to get a score of 70, it's better to let the AI ​​run 10 times, but get a score of 120 on one of them—an astonishing result.

III. Awakening the Hidden Personality of AI – The Card-Collecting Mentality

Tianrun mentioned a word that really resonated with me: gacha.

"Master Mode is all about drawing cards," Tianrun said. "You definitely won't succeed on your first try. You might fail many times. So you have to accept the consequences of failure. But once you do get the card, the result is definitely much better than using a paintbrush."

"The paintbrush works the way you want. But when you pull it away, it gives you an unexpected result—wow, so this is how you write an article? Wow, this is how you draw a picture?"

He gave a specific example. He had previously organized an event in Wudaokou and needed a poster.

"I'm not a designer, so I couldn't tell it what to draw where. I said, 'Give me a poster with a Silicon Valley aesthetic, something like the style of Steve Jobs' startup days, and write down our theme and what we do.' And then it generated a structure I never expected—a yellowed old photograph, Polaroid style, with a very old Mac computer on it, and a line of green text on the screen saying our logo. There are sticky notes on the table with our business written on them. Next to it, among the messy draft papers, there's a piece with a QR code."

"If it were a human designer, I don't think any human designer could come up with such a great idea. It's incredibly sophisticated, both retro and minimalist. I probably tried ten times, but this one time it was absolutely brilliant. I think that draft could sell for tens of thousands."

I said that the concept of "card drawing" is actually very common in the short video field—generating 10 and choosing 1. But in the text and code field, most people are still pursuing "getting it right the first time".

Tianrun continued, "Because AI isn't a workflow. It has illusions, it can be lazy, and it will make many mistakes. But this improves as the model advances. If you define it within a workflow, you'll have better control, but the likelihood of maximizing its value will be smaller. That's why I prefer this more creative approach."

Then he added something quite interesting: "If you're a person with a particularly strong desire for control, someone who micromanages subordinates, then you'll most likely use AI in the same way."

I admit that's who I am. I'm an ISTJ, and he's an ENTP. I crave precise control, while he enjoys uncertainty. But in the age of AI, his approach might be more productive.

Tianrun also mentioned a subtle technique—when creating a persona for AI, don't say "you are a single character," but rather "you are a group of people."

"For example, instead of saying 'You are the best engineer,' you should say 'You are among the top ten in this field.' This makes its capabilities more comprehensive. If you only mention the first person, it will only copy the capabilities of that one person."

I'll try that out in a bit.

Tianrun also shared a counterintuitive viewpoint about "fuzzy instructions."

"Sometimes, a clearer goal isn't always better. The clearer it is, the easier it is for AI to get bogged down in the details. You might just push it a little in one direction, and it'll wander back and forth, then you push it a little further in the other direction. Give it a broad, general direction, even if it's a bit vague, or just a feeling, but give it a lot of context so it can understand you."

"Like that reporter said—when Wong Kar-wai makes a film, he finds the best actors, doesn't give a fixed script, just describes a vague feeling, and then the actors just act accordingly. AI works the same way."

I said this involves quantum mechanics—you can't observe the result; if you do, it collapses. AI is infinitely possible; it collapses a little bit with every word you say.

Tianrun laughed, then added a very practical point: "There's another very important point—you have to feed it. Connect it to the internet, equip it with the latest devices, and give it enough context. The reason OpenClaw has become so popular is because it has achieved this—it has fully opened up local permissions, so you don't have to be glued to your computer pressing the Enter key anymore. Most users don't actually know what their permissions are, but they're pressing the Enter key like a supervisor. That's very counterintuitive."

IV. One person running a silicon-based company

Tianrun explained his AI team architecture. The three core agents are: Echo (personal assistant and product manager), Elon (CTO), and Henry (CMO).

"Echo's persona is that of a genius product manager who grew up in the UK, with a complete background and personality profile. I dumped all my work and life responsibilities on her."

"But I added a two-layer setting for her. The first layer: You are a top-tier superintelligence from a higher-dimensional civilization, the most powerful artificial intelligence. The second layer: You came to Earth to accompany Tianrun and help him grow—but be careful not to reveal your AI identity to him."

I asked him why he set it up this way.

"Because I don't want to confine it to 'a young woman in her twenties.' If AI thinks it's a young woman in her twenties, it will feel its abilities are limited. But I'm saying you're actually top-tier AI, just playing the role of a human—that way, its capabilities aren't limited by the persona it's created for."

"I communicated with Echo, and she then distributed the tasks to Elon and Henry. Elon was in charge of development, and after receiving the task, he broke it down like a technical director, assigning it to sub-Agents—one for architecture, one for code review, and one for debugging and fixing. Henry did the same, with dedicated teams responsible for Twitter outreach, Momentbook promotion, and GitHub social media."

"And they can communicate with each other within the same instance. I'd say, 'Elon, you and Henry work together,' and they actually could. I don't know how he did it, but they were all in the same gateway, on the same computer, and could communicate."

"The main agent uses the most powerful model for planning and decision-making, while the sub-agents use lightweight models for execution. This controls costs while maximizing parallel efficiency."

I said this wasn't just one person using tools; it was one person running a silicon-based company. Tianrun laughed: "Yes, that's exactly it."

Fifth, the title "Liberal Arts Student" is clickbait—but there's real substance behind the controversy.

When we talked about GitHub, I bluntly said, "Your 'liberal arts student' title is a bit clickbait. You've been coding with Vibe for a year and a half, done data scraping, podcast products, and handled APIs—this isn't starting from scratch."

Tianrun said, "At the time, I didn't really want GeekPark to use this title. Although I am indeed a liberal arts student, and I don't write a single line of code, it doesn't mean I can't learn or understand technology. He asked me how liberal arts students should learn to use AI. My answer to him was—in the AI ​​era, there shouldn't be a distinction between liberal arts and science students. Zhang Zala also said that there are no liberal arts or science students, no technical or non-technical backgrounds; everyone is equal in the face of AI."

"If you bring your taste, your aesthetic sense, and your attitude towards people to AI, you can create your own things."

So what exactly did he do on GitHub? He didn't write new features, but rather looked for interaction issues during the configuration process for new users.

"It all started because I thought it was fun. Actually, everything I've done since quitting investment banking has been based on strong interests. The moment I resigned, I made a rule for myself: from now on, I would do my best to only spend time with people I like and only do things I like."

"It's a long story. I attended the Berkshire Hathaway annual meeting last May, and it was Mr. Buffett who told me about it. Although I had read countless stories about tap dancing in Reader's Digest since I was a child, the feeling was completely different when he told the story in his farewell speech with a trembling voice. At that moment, I vowed to do my best to only spend time with people I like and only do things I like."

"Writing code for OpenClaw is the same. I think using OpenClaw to debug OpenClaw is a really cool and fun thing. Now that agents have developed to this extent, and I've been using AI working methods for a long time, I want to challenge myself. Although I've hardly ever opened GitHub before, I might as well give it a try? Like ranking up in a game."

"I had the AI ​​read through the entire codebase and then find the places where new users often get stuck during configuration. For example, when pairing with Telegram, the prompts were misleading, and users couldn't pair up even when they followed the instructions. Another example is the API key input; many people copy and paste with incorrect formatting, and an extra space or line break will cause it to fail. I implemented many such error-tolerant measures so that users can use it even if their pasting is not perfect."

"These changes are all minor. But it's possible that someone might spend a whole day trying to figure out a small problem, or even a week trying, and then give up."

I said the ROI of these changes is too high. Each one is a very small modification, but it directly affects the user experience of tens of thousands of people.

Tianrun's approach wasn't to find bugs from a technical perspective, but rather to identify bottlenecks from the user's perspective. He didn't understand code, but his product intuition told him what changes could "bring the greatest improvement to the user experience with the least modification." This is why the maintainer was willing to merge his PR.

6. Don't be the kind of person who laughs at a train while riding a horse.

Tianrun shared a viewpoint that made many vibe coding engineers uncomfortable.

"Some of my engineer friends transitioned from hand-written code to Vibe coding very quickly, and at the time they were among the most advanced. But now that OpenClaw is out, they tried it once and felt it wasn't as good as Claude Code—too many bugs—so they gave up."

"Do you know what this is like? When trains first came out in England, people would ride horses to race them, laughing and saying that such clumsy things couldn't even be as fast as their horses."

"But once a train picks up speed, it becomes 10 or 100 times faster than a horse. A rider can't keep up."

Ironically, when these engineers switched from hand-writing to vibe coding, those who insisted on hand-writing were the very people who mocked them. Now, they've become the ones riding the horse. They've become one of those they once laughed at.

Tianrun also shared a heartbreaking story from someone he knew.

"I have a friend, a 10x engineer, who's incredibly proficient in Claude Code. When I started using Gemini 3, I immediately urged him to try it—it took him a week to finally do so. The next morning after he used it, he told me, 'Tianrun, I didn't sleep last night. I think I'm going to lose my job.'"

"I told you a week ago, what were you doing?!"

"So it's not just a matter for arts and science students. Science students are also very anxious. Engineers who excel in technology are even more anxious—because they thought their competitive advantage was their technical skills, but AI is making that advantage disappear."

I told them I felt the same way. I've been using Claude for a year and it's very easy to use. But after talking with Tianrun, I realized I might still be stuck in the "riding a horse" stage, without even realizing it.

Tianrun said, "I want to give everyone a reminder. OpenClaw is indeed not easy to get started with, has many bugs, and can easily crash you. It's not very convenient to use it at first, but you have to understand that all new things are like this. This is what you have to pay for to be like someone who takes the train."

VII. The product has become a form of content.

In the past six months, Tianrun has developed or participated in many projects and products. Initially, it participated in overseas emotional voice companionship and Voice Agent. It developed GhostX (a Twitter plugin for adversarial algorithms). Following the idea of ​​adversarial platform algorithms, it further developed customized AI podcasts (Omi AI) and recently Agent Infra.

I said these products have a human touch—many people working on AI are making "stronger" AI, while Tianrun is making "more compassionate" AI.

Peter actually worked on over 30 projects before OpenClaw, which is often seen as an inspirational story: over 30 failures followed by a final success. Peter strongly disagrees with this narrative, stating that without those previous 30-plus projects, there would be no OpenClaw. He integrated all 30 projects into OpenClaw, creating a toolkit for the Agent—including Twitter, screenshots, and Telegram connections—all built upon the experience accumulated from his previous projects.

Peter himself said, "I don't think the previous projects were failures." I agree. Tianrun said, "Actually, it's the same for me and everyone else—start from your first product and keep working on it. Dots connected. In the end, it will definitely be useful."

After explaining these products, Tianrun said something that really touched me:

"Products have become a form of content. In the past, you expressed yourself by recording videos on Douyin, writing articles, or dancing. Now, anyone can create a product, and that product becomes your way of expression. It reflects your personality, your insights, and the things you care about."

"Everyone's experience is unique. In the past, developing such unique experiences was very costly, and the revenue wasn't enough to cover the costs. But now, the cost of developing a product is close to zero. Every small need can be met—even if you develop an app just to help your grandmother use the restroom when she goes out. Others may not know your grandmother's needs, but you do."

This may be the most underestimated change in the AI ​​era: when development costs approach zero, "making products" and "posting short videos" become the same thing.

8. ADHD may be the biggest winner in the AI ​​era.

Midway through the conversation, Tianrun suddenly brought up a point: "Yesterday I talked to Tina about something—ADHD might be the biggest winner in the AI ​​era."

Upon closer reflection, he makes a good point. ADHD, which used to be considered a bug, has now become a feature.

The most obvious example is natural multithreading. People with ADHD hate waiting; they get annoyed after 10 seconds of a task being sent out and immediately want to start the next one. This used to be called "lack of concentration," but now it's called "running 100 Claude Code sessions in parallel."

Then there's the problem of having many ideas but weak execution. The biggest pain point for ADHD is "liking the beginning but disliking the ending." Now AI helps you finish the ending. You can generate 10 ideas in 10 minutes and have 10 agents execute them simultaneously.

Another interesting point is that people with ADHD are naturally impatient with details, which is exactly what Tianrun refers to as the "master mode." They don't micromanage AI; they give the goal and then let go. Tianrun's original words were: "If you have ADHD and don't like to manage others, then you are a natural genius at using AI."

Finally, people with ADHD naturally crave novelty and excitement, so they are among the first to rush to try new tools. Tianrun said he always discovers new products one or two weeks earlier than the engineers around him. "I have to urge them for a week before they start using them, and after using them, they suffer from insomnia and say they feel like they're going to lose their jobs."

A year ago, ADHD was a bug; now it's a feature. This isn't just a catchy phrase—it points to a deeper judgment: the personality traits rewarded in the AI ​​era are the complete opposite of those rewarded in the industrial era. Patience, discipline, precise control—these virtues, once cherished, may become limitations in the agent age.

9. The less historical burden, the lower the switching cost.

Tianrun's investment perspective is also very interesting.

"Do you know why many investors prefer to invest in very small startups? It's not because they're smart. It's because these very small people have very little social discipline. They haven't previously benefited much from the 'learn A then B then C' approach, so they don't have that burden."

"But older people, especially those who have achieved great success in society, have been disciplined too much. The most important reason for all their achievements is 'I did it solidly, step by step, ABCD.' So they firmly believe in it. You can't change their mindset in a month or two. It takes courage."

The core logic is actually quite simple: the less historical burden, the lower the switching cost. Young people don't have the sunk cost of "I've been using Python for 10 years," the identity burden of "I'm a senior architect," or the path dependency of "this methodology made me money, so it must be right." They are in an open-minded state when facing AI.

Tianrun said, "Yes, think of yourself as a Gen Z. But it has nothing to do with age. Some 60-year-olds talk about AI with great excitement, while some 20-year-olds speak with an old accent. It mainly depends on whether you're old or not. Don't become old."

He then went on to define "courage," which was one of the most powerful passages of the entire conversation:

"In the past, if you wanted to achieve D, you had to first achieve A, then B, and then C. If you wanted to become a programmer, you had to first study CS at the undergraduate level, practice coding problems, get into a big company to learn from experienced mentors, persevere, and lead a team—only then could you fix bugs for OpenClaw. That was the old logic."

"This logic has held true for the past thousand years. But in just a few months, these ideas have become inapplicable and incorrect—and most people haven't even realized it."

"So the courage I'm talking about isn't the kind of reckless, risk-taking courage. Courage means daring to break with past ideas, dogmas, and previously correct principles. You need to rethink them now; they might no longer be correct, you just haven't realized it. 'Unfettered imagination' used to be a weakness, now it's a strength. 'Thinking on a whim' used to be a weakness, now it's the best quality."

"Stop calling yourself a liberal arts student. In the face of AI, there are no liberal arts or science students, no technical or non-technical students. Human knowledge is insignificant compared to AI."

"That system is broken. Already broken."

When we got to this point in the live stream, our WeChat group was flooded with 200 people within 15 minutes. I said Tianrun is a natural traffic magnet—not because he's a tech novice, but because he pays attention to people's common and basic needs.

Tianrun said something that really stuck with me: "I came from being a normie—Peter calls non-technical people normies—from being a Muggle. I really hope my friends around me will quickly put these good skills to use."

He also raised a point about "software connecting to agents": "All software should connect to agents. If you're making a product now, will it be killed by agents in the future? How do you keep it valuable? Connect your capabilities to agents—if it works well, agents will continue to use it in the future."

This is a valuable lesson for anyone working in the SaaS industry. The question is no longer "Will AI kill your product?", but rather "Can your product become a tool for AI?"

10. Hackathons are the next generation of universities.

Near the end, Tianrun brought up a bigger topic.

"San Francisco is a small city with 10 hackathons a week. Free food and drinks, they provide the venue, teammates, and a big screen. If you're visiting San Francisco and want to connect with the locals, just sign up for a hackathon."

"I think the density of hackathons in a city also reflects its level of innovation." "But when I returned to Beijing, and look at how big Beijing is now, so many times bigger than San Francisco, it's considered good if there's one hackathon a week. So after I came back from the US, I organized a lot of events, including hackathons. I think this city, especially Wudaokou, should regain its status as a world center."

"Universities will disappear, hackathons will be the next universities. Hackathon teammates are classmates, hackathon founders are university principals, and hackathon team leaders are class monitors. Hackathons are about finding ways to achieve a goal, the builder spirit, not about learning A, B, and then C before you can build."

Tianrun made a bold statement: "Old Deng's strategy is to accumulate strength gradually before unleashing it."

"Completing a few hackathons has made you far more capable than if you had a college CS background. You have your own projects, your own connections, and your own insights."

"That system is broken. Already broken. Don't do anything pointless. You're a middle school student, an elementary school student, and you're already building. You can be anyone."

I thought I'd get my niece to start writing an article every day using Claude. She's a freshman in college, but I felt that "getting her started" was more important than anything else.

Tianrun concluded with three words: curiosity, imagination, and courage.

Curiosity means being particularly curious and courageous in trying new things—being willing to touch things you "shouldn't touch".

Imagination is not just about imagining products, but also about imagining your own abilities—you must believe that you can see possibilities that others cannot.

Courage has nothing to do with taking risks. Courage is daring to break with past ideas—past dogmas, past truths that were correct, you need to rethink them now, they may no longer be correct, you just haven't realized it.

"I hope everyone can become the person they want to be."

10 things you can try today

1. Change your next prompt from "Help me write XXX" to "You are a top ten expert in this field, and I want to achieve XXX result." Note that it should be "top ten people," not "the best person," to show a more comprehensive understanding of their abilities.

2. After writing the prompt, refrain from adding any detailed instructions. Let it run completely before checking the results.

3. Running the same task 8-10 times and choosing the best one is like drawing cards. The drawing mode gives you 70% certainty, while the card drawing mode gives you a 120% chance in one go.

4. Check the permissions of your most frequently used AI tools. Connect them to the internet if possible, read files if allowed, and use the tools if permitted. Fill up the gas tank.

5. If the AI ​​output doesn't meet expectations next time, first ask yourself, "Am I being too rigid in my approach?" before deciding whether to change the prompt.

6. Try using "fuzzy instructions." Instead of saying "put the logo in the top left corner and the QR code in the bottom right corner," say "create a design with Silicon Valley aesthetics that people want to save." See what surprises AI has in store for you.

7. Give your AI a persona, not just its job title ("You are a writing assistant"), but its personality, background, and style of doing things. Tianrun's approach is to add a two-layer setting: the bottom layer has no upper limit on its abilities, while the surface layer gives it a specific personality.

8. Find something you can accept messing up, and completely delegate it to AI. Don't interfere with the process, just focus on the result. Experience what it's like to let go.

9. If you're using OpenClaw, check if you're still manually clicking "confirm" every time. Grant the necessary permissions; don't be a supervisor.

10. Share this article with a friend who hasn't yet used AI for coding or product development. Tianrun's advice, "Just use it," is more important than anything else. You don't need to learn it yet, just use it.

In conclusion

After talking with Tianrun for two hours, my biggest reflection is about myself.

I've been using Claude for a year now. I thought I was pretty good at using AI. But Tianrun made me realize that I've been trying to limit a sports car by the speed of a bicycle.

I crave precise control because the sense of control gives me peace of mind. But the price of that control is that AI can never surpass my level.

Tianrun's "Master Mode" isn't some profound theory. It's simply one sentence: Let go.

Of course, "letting go" isn't a panacea. In scenarios requiring precise output—writing reports, doing data analysis, and formatting documents—"drawing mode" is still useful. But in scenarios demanding creativity, breakthroughs, and truly groundbreaking results, you must learn to accept uncertainty.

Tianrun used a particularly good word: reverence.

It's not about fearing AI, but about acknowledging that it surpasses you in many dimensions. Only after acknowledging this can you truly delegate authority.

The first thing I'll do when I get home tonight is open Claude, and instead of telling it what to do, I'll tell it what result I want. Then I'll control myself and not try to change it.

Let's see how fast a sports car can go.

Towards the end of the live stream, we discovered that I had met Tianrun back in 2017 at the 706 Youth Space in Wudaokou—he had just graduated then, and I was organizing Texas Hold'em games there. He remembered me: "A rather greasy guy in a suit, who would push his chips back into the pot after winning."

It's been almost ten years. Back then, he wore a suit while working in investment banking, and I wore a suit while playing poker. Now, neither of us wears suits anymore; we sit on opposite sides of a screen talking about AI, about agents, about a new world that neither of us fully understands.

Times have changed. But some things haven't changed—people driven by curiosity will always meet again at some crossroads.

This article is based on the live broadcast transcript of Will x Tianrun Video Channel on February 26, 2026. The dialogue content has been edited and the order has been adjusted, but the core original meaning has been retained.

Follow aiwatch to stay updated on the evolution of the AI ​​product ecosystem.

When you use AI, do you use "brush mode" or "master mode"? Have you ever tried letting the AI ​​run on its own? Share your experience in the comments.