Morning, CEO!
Today we’re looking at Kaijie Chen. He’s 26. He’s on his third startup. His latest creation, Macaron AI, racked up 300,000 users and churned out 200,000 user-generated apps in just a few months.
Meanwhile, I’m over here celebrating because I remembered to cancel a free trial before my credit card got hit. We are not the same.
But that’s okay. We don’t need to be the 26-year-old prodigy; we just need to ruthlessly steal his strategy for our own operations. Let’s look at how he’s building a personalized AI empire—and how we can use his logic to upgrade “Me Inc.”
1. Stop Reading the Manual, Start Whacking the Ball
Here is the biggest trap I fall into—and I bet you do too. I call it “Procrastination by Preparation.”
We read the books. We watch the tutorials. We consume the strategy. We feel productive because our brains are filling up with data.
Kaijie Chen calls this the “Pre-training Era.” It’s how early AI models were built. You feed the machine the entire internet, and hope it gets smart. But in 2025? We’ve run out of internet. The data is exhausted. Reading more isn’t making the models smarter.
So, Chen pivoted to the “Era of Experience.”
He uses this tennis analogy that hit me right in the gut:
You can watch 10 hours of Roger Federer playing tennis. You will intellectually understand the game.
But if you pick up a racquet and swing once? Even if you miss? That single second of failure contains more data than the 10 hours of watching.
Your arm feels the weight. Your eyes judge the depth. Your brain gets an immediate “Error! Adjust Left!” signal.
This is Reinforcement Learning (RL). It’s learning through the feedback loop of doing, failing, and adjusting.
The Takeaway for Us:
Your “Agency of One” is stalling because you’re still pre-training. You’re over-consuming information.
Chen’s entire tech stack is built on High Information Gain. He doesn’t want more data; he wants better data. The only way to get that is to enter the arena.
Stop researching the perfect cold email strategy. Send ten terrible emails today. The rejection you get is the “Reward Model” signal you need to optimize the next ten.
Stop planning the perfect project roadmap. Ship the messy beta.
The market provides the “RL,” but only if you provide the swing.
2. Your “Vibe” is the Moat (Product Taste)
We live in a world where everyone has access to the same “Intellectual Capital.”
You have ChatGPT. Your competitor has ChatGPT. I have ChatGPT (and a mild caffeine addiction).
If the raw intelligence is commoditized, why does anyone pick Macaron AI over the giants like OpenAI or Google?
Taste.
Chen realized that most AI tools are just cold, efficient assistants. They sit there, blinking, waiting for a command. They are reactive.
He built Macaron to be Proactive, Dynamic, and Vibrant.
If you tell a standard AI you’re sad, it lists three local therapists. Efficient? Yes. Cold? Absolutely.
If you tell Macaron you’re sad, it remembers that you like hiking. It proactively suggests a trail. It checks in on you the next morning. It acts like a friend, not a search engine.
This is what Chen calls “Product Taste.” It’s the difference between a tool and a partner.
The Takeaway for Us:
In your professional life, your “Deliverables” are likely commodities.
The report you write? AI can do it. The code you ship? AI can generate it. The strategy deck? AI can outline it.
If you compete on utility, you lose.
You have to compete on Taste and Proactivity.
Don’t be the freelancer who waits for the client to ask for the status update. Be the partner who anticipates the anxiety and sends the update before they ask.
Don’t just deliver the spreadsheet; deliver the insight about what the numbers mean for their specific goals.
Your “Agency of One” needs a personality. It needs to be opinionated. It needs to care.
Intelligence is cheap. “Giving a damn” is the premium feature.
3. Don’t Just Do Work; Build Systems That Do Work
This is the part that blew my mind.
Macaron isn’t just a chatbot. It’s a factory.
Users don’t just chat with it; they tell it to build “Mini-Apps.”
A user says, “I want to track my calories by taking photos of my food.” Boom. Macaron writes the code and generates a mini-app for that specific purpose.
A user says, “I need to analyze my spending to save for a house.” Boom. A finance tracking app appears.
Chen isn’t trying to build every feature himself. He built a system that allows non-technical users to build their own tools.
He has 300,000 users, but they’ve created 200,000 apps. That is a staggering ratio of creation.
He is moving from “Creation” (him building the app) to “Usage” (users building the value).
The Takeaway for Us:
We need to stop treating AI as a typewriter that writes faster, and start treating it as a developer that builds tools.
If you find yourself doing the same task three times, you are failing as a CEO.
Don’t write the email response.
Build the prompt chain that analyzes the incoming email and drafts the response in your voice.
Don’t manually summarize the meeting notes.
Build the workflow that records, transcribes, summarizes, and emails the action items to the client automatically.
You are not the worker bee. You are the architect of the hive.
Chen’s users are building “Mini-Apps” to solve their life problems. You need to build “Mini-Workflows” to solve your business problems.
Build the asset once. Let it run forever.
To Recap:
Stop watching tennis. The only data that matters comes from swinging the racquet and hitting the net. Ship the work.
Vibe check yourself. Your utility is replaceable. Your “Taste” and proactivity are not. Be a partner, not a tool.
Build the factory. Don’t just do the task. Build the mini-app (workflow) that does the task for you next time.
Kaijie Chen is betting that AI’s future isn’t about bigger brains, but about better experience.
I think he’s right.
Now I’m going to go ask an AI to build me a mini-app that reminds me to stop doom-scrolling and go to bed.
Links:
https://x.com/KaijieChen12236
https://macaron.im
https://36kr.com/p/2610259787456641
https://36kr.com/p/3559477201484678
https://mp.weixin.qq.com/s/mGYTK8s4fjBm46zh703acg
