For the past two years, Zhang Yueguang seemed to have disappeared.
This product manager created a phenomenal AI product like Miaoya in 2023, but just 3 months after Miaoya became popular, he resigned from Alibaba. He said this stemmed from a night he spent sitting alone by the Kamo River in Kyoto, Japan, where he gained clarity on many important life decisions.
In early 2024, Zhang Yueguang raised funds to start Muyan Zhiyu. This company became one of the highest-funded AI application startups that year.
Some say investors "blindly bet" on him for 4 rounds without a product, accumulating nearly 300 million RMB. Others joke that Zhang Yueguang is running his company on the interest from his funding.
Over the past two years, I've met Zhang Yueguang many times. The product forms he was experimenting with changed each time, and he didn't hide his confusion, frankly admitting that for a while he "couldn't understand." The turning point came half a year ago.
**What kind of trial and error and exploration does a product manager, trained in the internet era, go through when he shifts to AI? What cognitive gaps must he bridge? — To my great surprise, Zhang Yueguang said he only later realized that Miaoya was not an AI Native product at all.**
Muyan Zhiyu is currently working on two main products: the AI otome game "Star Sleep" and the upcoming Agent product "Dokie." He says his exploration phase is over, and he has finally found what he is willing to commit the company's full weight to.
What follows is my interview with Zhang Yueguang.
Looking forward to 2026, when we progress together with AI!
(Zhang Yueguang thinking about life by the Kamo River during a long night)
**02:00 Guest Biography**
Graduated from Tsinghua Computer Science in 2012, discovered in junior year that he couldn't compete in coding.
First failed startup and product (Yuanyin) in 2018.
Was very confused by the end of the last startup: didn't know why he started it.
My two experiences at ByteDance (Time Album, Toutiao Special Projects).
My career path before this startup: SaaS company > Alipay > ByteDance > Startup > Alibaba Youku.
**In the early hours of National Day 2023, sitting by the Kamo River in Kyoto, decided on the second startup.**
My principle: "I want to do the same thing with the same group of people from age 35-45."
**16:56 Playing the game of leveling up in big tech companies**
At 27, he was ByteDance 3-2; at 30, he was Alibaba P9.
Big tech companies generally have two types of battles: external battles and internal battles.
**Frankly, the big tech game turns most people into internal combatants.**
AI provides an increment for external games; otherwise, one can only play internal games.
The best reasons to start a business are two: 1. You just want to be the boss; 2. You really want something to happen; if you want to get rich or famous, the expected return from starting a business is very poor.
Long-form video platforms are not platforms, just as cinemas are not platforms; ultimately, they are content products.
In 2023, I said in a meeting: "From today onwards, I will not participate in any meetings about Youku."
**28:28 Miaoya is not an AI Native product, so what is?**
I like One Way Door products.
Product design in the internet era is process-oriented design; good products like WeChat are amusement parks, but with limited freedom.
But AI Native products are very open, not only open in input but also open in output.
**From internet product design to AI Native product design, there's a huge paradigm shift: from process design to context design.**
The traditional collaboration methods of internet product teams are very likely to encounter problems today.
**42:04 What kind of AI organization is needed to design AI products? Decision boundaries are blurring, Taste becomes very important, shifting from linear work to a two-stage process.**
Miaoya and Meitu Xiuxiu are not in the same league; it's essentially a portrait photography business.
We found it difficult to make one model achieve realism, likeness, and beauty, so we used three models simultaneously.
At its peak, it used nearly 20,000 inference cards.
Miaoya added a happy moment to my life: the day my daily revenue exceeded 1 million RMB while drinking with colleagues at Liangmaqiao.
From July to October 2023, I was incredibly busy; in October, I took a vacation and decided to resign. Everything came to an abrupt halt.
**01:01:16 Embarking on the second startup journey**
This generation of AI technology should be about "creating AI populations," not "creating AI services."
AI will be connected to the entire human network.
Today's professional skills are depreciating; diverse tastes and broad knowledge determine the team's ceiling.
"Humans should be responsible for Will, humans should be responsible for Skill."
As one of the highest-funded AI application companies in 2023, why did investors "blindly bet" without a product?
People in the industry joke, "We're spending interest every day."
**01:14:29 Multiple product choices and changes after hitting the road**
A definite path: AI otome game with interaction as the core (Star Sleep).
An uncertain path: spent 1 year exploring, made three or four small products, such as podcasts, information tools, image tools.
Let's talk about why we previously made a podcast product and then abandoned it?
The only opportunity for new content platforms: new media, new interaction, creating new platforms.
**Sora can never become a new platform, but ChatGPT can.**
**I think the commercial space for Chatbots is unimaginably large, possibly a business model never seen before in human history—it hides the ability for long-range decision control.**
**Douyin-style short video interaction deprives you of informational autonomy; Chatbot has the opportunity to deprive you of autonomous decision-making.**
It can control you, so it can take a very high Take Rate (commission rate).
**02:01:21 Transitioning from a model paradigm to an Agent paradigm**
Let Agents do more things you absolutely cannot do.
Manus is a "Silicon Valley fairy tale more Silicon Valley than Silicon Valley"; Meta bought a representative of a paradigm.
**I don't want to make an Agent like Manus; new product to be released soon: Dokie.**
I care a lot about Latency (delay) and hope for continuous collaboration.
Why did Dokie choose such a narrow niche as PPT? — "Don't talk about 'the first,' we're starting 'the last'."
My product philosophy: not "the first," 90 points equals 0 points, no one thinks this is made by AI, create a One Way Door / single-door experience.
Answering the "Heavenly Questions": How do you view competition with model companies? (Application companies are all using Claude's tool call).
Star Sleep is not about my AI friend; Dokie is about an AI friend that breaks the boundaries of my abilities.
[More information]
Disclaimer: This content is not investment advice.
Original title:
130. 张月光创业两年首次访谈：妙鸭不是AI Native产品、流程到上下文设计、One Way Door和乙女游戏
Original description:
<figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>过去两年，张月光…
