---
title: "From PMF to MPF: The Underlying Logic AI Founders Must Replace"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-04-20T10:01:48+00:00"
canonical: "https://ffcap.cn/en/research/src-20260420-01html"
source: "https://uniqueresearch.substack.com/p/src-20260420-01html"
language: "en"
---

# From PMF to MPF: The Underlying Logic AI Founders Must Replace

_Original · Unique Research · 2026-04-20_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, repeated examples and full panel, including all named speaking turns and their continuation paragraphs. Product, user, market and technical figures are source or speaker claims, not independently audited findings. Company, personal and product names are transliterated where official English forms remain unverified. "PMF" (Product-Market Fit) and "MPF" (Market-Product Fit) are preserved as source terminology. The source is dated April 20, 2026._

Unique Awards

The Truth About AI Going Global: You Think What You Lack Is a Moat, But Actually It's Traffic

In AI going global, what many teams truly lose is not that their technology isn't strong enough, but that they have already fired all their bullets before being seen.

"

One team spends three months grinding onunderlying code, and after launch no one cares; another founder hasn't even written code, just put up a minimal "fake page," but pulled in hundreds of real registered users with one tweet—where is the gap?

At Unique Awards · Hangzhou AI WEEK, a roundtable on "Globalization of AI Productivity Tools" pierced a truth that makes domestic tech guys very uncomfortable—in the context of AI going global, the "technical barrier" you are proud of may be worthless at all. Four AI founders who have sold products to most countries around the world gathered together and gave a set of extremely counterintuitive survival rules.

Stop Coding in Silence, Go Get Traffic First

If you were to make a global-facing AI tool, what would you do first? Most teams would say: read papers, tune models, write code, build features.

"This is a big mistake." Steven, co-founder of ChartGen AI, directly called it out.

But in today's world where Web Coding tools are everywhere, writing a "usable product" is no longer a threshold at all. Stevenraised an even crueler conclusion: the logic has now shifted from the familiar "Product Market Fit" to "Market Product Fit."

You have to first prove to yourself that there are fish in this pond, and then go build that expensive net.

"Many Silicon Valley founders spend 60% of their time every day promoting themselves on LinkedIn and X." Steven revealed that in this so-called era of "AI Wrappers" running everywhere, how you let your tool be seen by real living people is far more important than whether your architecture is awesome.

Xu Zuobiao, founder of Dynal.ai, is also equally fierce and direct. His team makes AI tools to help people "compete" on personal image and posting on North American LinkedIn. "When people ask you what your payment rate and retention rate are, it's all nonsense, because you have no traffic at all." Xu Zuobiao spoke very plainly, "If you have 100,000 visits every day, you can always keep some people to make adjustments for you—that is PMF."

While you are stilltorn about how to fix bugs more perfectly, others are already grabbing customers with a few screenshots. This is called dimensional-reduction strike.

When AI Becomes Infrastructure, Where Is the Moat?

If you have traffic, the next question is: the competition among large models is so fierce—today the feature youpieced together with prompts, tomorrow OpenAI or Claude will wipe it out with a major version update. Where is the moat?

A wave of extremely dense viewpoint collisions emerged on this topic across the board.

Ziwen, co-founder of AirJelly, has a relatively "stubborn" viewpoint: the moat is in the hard-core strength of the product itself. There are various open-source or free Agents on the market (such as OpenCodeseize Claude's ecosystem), but he feels that good open-source can instead enlarge the plate and publiccognition, and essentially it is still a head-to-head battle of tool experience.

But Steven gave a completely different and extremely pragmatic approach: don't look for barriers in the AI function itself—go to the backend for asymmetric data advantages.

"As AI gets hotter and hotter, the moat of the function itself will get lower and lower, almost flattened." Steven hit the nail on the head. Everyone is tuning similar models—why should you be stronger than others? ChartGen's path is to connect to those high-quality Data APIs (such as Nasdaq's exclusive data). When users use their open-source Skill to draw very beautiful dashboards, what truly needs to be paid for and locked in is actually the underlying high-density data source that ordinary people cannot get.

Longyi, founder & CEO of Seede AI, pointed out a very interesting new direction—selling "non-standard products."

What they do is use large models to help ordinary people get graphic design done. "If you input extremely simple, minimal or cyberpunk prompts, what the large model spits out often tends toward 'averageness.'" But human aesthetics pursues eternal freshness and extreme variation. So their moat is in providing "new aesthetic Context."

When technology is equalized, what is truly valuable comes down to two things:rigid demand data sources that others cannot get, and taste that cannot be derived by formulas.

Day 1 Global, What Exactly Are You Globalizing?

The four products on stage have vastly different forms, but they share an extremely eye-catching common point: Day 1 (from the first day of founding) they were aimed at the global market.

Why not first penetrate the large domestic market? Is it because domestic is "too competitive" and they want to take refuge overseas?

In fact, globalization is not escaping domestic competition, but using global differentiated dividends tocrazy put "leverage" on the product.

Steven told a fascinating detail: on the 10th day after their product launched, users from 110 countries around the world flooded in. Because the user base was large enough and distributed widely enough, they immediately discovered that demand in the distant Middle East and North Africa region was surprisingly similar, and could quickly do regional-leveltargeted version iteration.

Because the core desire of "productivity" is highly synchronized across all humanity. "Everyone uses AI products about the same, just like everyone uses Office." Longyi summarized it this way. Although there are subtle differences in cultural context, the underlying skeleton leads to all humanity.

But this by no means means you can sit in an office building in Wangjing and guess the pain points of American or Japanese workers out of thin air. "If you have theconditions, you must go deep into the local area." Ziwen strongly advised this batch of going-global entrepreneurs, "Really go to Japan and see what knowledge workers are actually doing, attend local AI events. If you don't chat with locals, what you get will always be distorted second-hand demand."

If the Large Model Gets 10x Stronger, Are You Afraid?

This is almost a nightmare lingering in the hearts of all AI founders now: I painstakingly built a set of automated workflows—if next year's large model can directly do it all in one go, am I directly out of a job?

"If you are truly building product service scenarios based on large models, when the model gets 10x stronger..." Xu Zuobiao paused and said steadily, "you would laugh awake from your dreams."

Because in the future, what limits you is no longer that meager computing power or the model's poor comprehension, but whether your scenario can catch those huge demands that once seemed "excessive and whimsical." As long as you solidly build the product and satisfy the scenario, the stronger the model, the easier your business will be.

AI is not here to take away your business—it is just a ruthless mirror. It can reflect your ultimate aesthetics, your keen business sense, and of course, mercilessly reflect those futile efforts of building cars behind closed doors.

More Conversation Details

Unique Awards · Hangzhou AI WEEK Trends Roundtable Panel

"From Open-Source Rooting to Global Growth: New Opportunities for the Globalization of AI Productivity Tools"

Guests:

Seede AI — Founder & CEO — Longyi

Dynal.ai — Founder — Xu Zuobiao

ChartGen AI — Co-founder — Steven

AirJelly — Co-founder — Ziwen

Moderator: EPIC Connector — Product Lead — Shawn

Shawn: Now let's start with Teacher Longyi, and in this order briefly introduce each company's product.

Longyi: My name is Longyi. The product we are making now is called Seede AI. It is actually a tool that uses large models to help ordinary users do graphic design. Our product already has several million users domestically, and recently in early March we also launched our overseas product, called Vis.

Xu Zuobiao: Our company is Dynal.ai, and this product is a new product for our going-global. It mainly does LinkedIn customer acquisition, helping professionals who want to acquire customers in North America or build influence there, within 30 to 90 days, build a very real, professional and sustainable personal image on LinkedIn. What users need to do is set a goal, then review the post content, and it can be published—roughly that's the situation.

Steven: Hello guests, my name is Steven, currently co-founder of ChartGen AI. Our product mainlytargeting going-global users, doing Data Agent in the data analysis field. Currently the product has two forms: one is a SaaS tool for normal users; at the same time we have also launched a Skilltargeting the OpenClaw ecosystem, allowing Agents to directly call our skills to do data visualization analysis, including some Dashboards. Our team was Day 1 global, basically with good user growth in North America, the Middle East and South America. Today it is also a great honor to share with you some of our experience and lessons in overseas growth.

Ziwen: Hello everyone, I am Ziwen. The product we make is AirJelly. AirJelly is a proactive context-aware assistant, with three main functions: first, we can感知 the user's screen and record what they do every day; second, we have a Proactive Agent—because we can感知 the screen, we can know the user's intent and provide proactive help at just the right time; third, we are also a general Agent that can help users execute most tasks that need to be done on a computer, such as writing documents, making PPT, etc. We are a company serving the world, Day 1 Global. AirJelly has recently started closed beta one after another, and we will陆续 send out invitations.

Shawn: Everyone is excellent global-facing products. Although the product forms are different and they serve different scenarios, the common feature is the global-facing positioning. But actually in this process there are differences: some products first did domestic verification, then did the overseas version; some are based on ecosystems familiar to overseas users like LinkedIn; or products derived from the open-source OpenClaw ecosystem. My question is: in the process of globalization, what is yourcognition of product positioning? Do you prioritize domestic verification first then globalize, or directly position as global from the beginning for development and promotion?

Longyi: Actually our Seede AI may be a bit different from the others here. We can say we first ran domestically—started running domestically last year, and only this year truly let the same product go overseas. But I personally believe our product and target users are themselves global. Why first run domestically? Because our product still needed some engineering polishing in the early stage. Fortunately, users' spontaneousspread quickly helped us accumulate users domestically, because we actually didn't do any promotion domestically. So it looks like we first did PMF domestically then went overseas, but this was just accidental. I firmly believe that a universal AI product can cross cultures—it's just that we need to do some specific customization in different cultural contexts, but the entire product structure itself should be global. Everyone uses AI products about the same, just like everyone uses Office.

Xu Zuobiao: Actually our product is our second going-global product; the previous one was for domestic scenarios. This new product was doing US-biased scenarios from day one, because at the time I stayed in the Bay Area for about 50+ days and found that every day I added many people's contact info, and their LinkedIn was all very "competitive." They may have posted hundreds of Posts, with very long text and various配图 photos; while I looked at my own—only four or five, each two or three sentences—the gap was huge. Every time I thought about posting on LinkedIn I got a headache, because I would think: is my idea valuable? Will others laugh at me? Is there traffic? Is the copy OK? Is the text-image strongly correlated? So I thought whether I could hit my own need to make a product. After chatting with many friends and finding they also needed it, I made it. We can upload a lot of materials to convert into high-quality Posts, and also spend a period of time building your Brand DNA, so that the content it posts is coherent and consistent, not making you look split. At the same time you can spend ten minutes and finish all the Posts for the next four weeks. We were actually Day 1 biased toward serving the US group.

Shawn: Then will you consider product design similar to the domestic ecosystem later?

Xu Zuobiao: No, we only do overseas.

Shawn: Because LinkedIn itself has a lot of user needs and practical applications, and the ecosystem is very good, focusing on it is also a relatively good strategy.

Xu Zuobiao: Maybe we will first finish LinkedIn, then do Twitter or Facebook.

Steven: For our company, being Day 1 Global has a very important reason: we want to pull up the product iteration speed. For example, on the 10th day after the product launched, there were users from about 110 countries around the world. We found that user needs in regions like the Middle East and North Africa are very similar. After accumulating a large number of users, we can do faster regional-level Localization. You can quickly do Feature Updates based on users' Queries—for example, if Japan and South Korea needs have common points, you can do targeted upgrades. Understanding the common demands of various regions around the world for data analysis in a short time brings relatively high leverage to the product. Of course, domestic AI is also very hot now, so we are also trying to release open-source versions of Skills on OpenClaw and domestic platforms, letting domestic users use basic capabilities for free; but we feel that in terms of payment capability, overseas C-end users are still stronger than domestic. So on one hand doing overall user growth overseas, and on the other hand releasing open-source tools domestically—this is the current strategy.

Ziwen: Our company chose Day 1 Global because we believe productivity tools are the same for workers around the world, and everyone's needs are actually similar. When doing Marketing, we learned that different regions have different usage habits and cultures, and through Day 1 Global we can reach real needs,反向 driving product iteration. For example, open-source things like OpenClaw are also Global, and good AI products cantargeting global users. We hope AirJelly will ultimately also be recognized by users around the world.

Shawn: In the AI field, China and the US are considered two leading players, with different product forms. Chinabiased向 open-source ecosystems, while the US like Claude can have good commercial returns through closed source. I特别 want to chat about: for open-source-based scenarios, what do you think is the core competitiveness or "moat"? Because open-source ecosystems, based on their free nature, may not be so easy to commercialize.

Ziwen: First, I think if there is an open-source product similar to AirJelly, it is definitely a good thing—everyone can enlarge the cake together and cultivate better user心智. Second, our closed-source competitiveness lies in the strength of the product itself. For example, there are many open-source Coding Agent products on the market, like OpenCode mayseize some Claude users, but because Claude's capability is strong enough, there are still many users willing to pay for it—open source just provides another choice. Ultimately it will make the number of people using this type of product larger and larger. More competitors entering is actually a good thing—they may use open source and then find AirJelly has better capability, and pay for us.

Steven: Open source itself is a very good marketing event, letting more people know us and enlarging the cake. But at the same time, what our closed-source version does more is data connection. For example, you make a very good Dashboard through our open-source Skill, but the high-quality data sources called (such as Nasdaq data) require API payment. In our view, as AI gets hotter and hotter, the moat of the function itself will get lower and lower, but your Source data quality is your core. The high-quality Data APIs we connect to at the backend are the barrier—you must pay to use them. So we are walking both paths: open source is a good way to acquire customers, while closed source builds a very thick barrier at the data level.

Shawn: Today we are discussing the globalization of AI productivity tools. How do you view the difference between "AI+" and "AI Native"? How do you define future product forms?

Xu Zuobiao: To put it more intuitively, it is "if you didn't have a large model, this thing would be废了." In the past, many products may have had high fees and large users, just adding AI translation or AI rewriting in some small places—this also has value. Take Adobe for example; maybe after a while there will be a company completely based on large models doing design, which may eat into its share; but some old email systems in the US, although hard to use, still have considerable annual revenue, so each has its own way. Our new product is definitely biased toward being based on large models, without the various problems of past graphical interfaces. Of course large models also have problems, such as occasionally uncontrollable processing of ultra-long context—you need to make trade-offs and choose a suitable scenario. As time goes on, the space and future for rebuilding new things based on large models are considerable.

Longyi: I started making Seede AI in 2024. Before that, at another company, I made website-building products through no-code thinking (visual drag-and-drop). Now you can hear about AI-based Web Coding website-building tools like Lobe, and you canobvious feel that they have a generational gap with no-code tools in technology and onboarding threshold. Existing engineering plus AI (traditional software) has a skeleton of traditional algorithms, and its biggest shortcoming is insufficient generalization capability. But current AI Native products have very strong generalization capability, because the model has cross-industry knowledge and can guess the user's intent based on context. There is a very big difference between traditional software and AI Native software, and the generational gap in future experience will get larger and larger.

Shawn: Next I hope everyone will share: in the process of globalization, what may be the biggest problem or cognitive difference point encountered? Can you give some thoughts for entrepreneurs who want to make global products later?

Steven: I feel this particularly deeply. We started going global last year. At first we spent a few months making the product, feeling technology was more important and Go to Market (GTM) was not so important—that was a big mistake. For example, on the foreign Product Hunt platform, there are about 400 to 500 products released every day. Now Web Coding tools are so hot that technology is no longer such a big barrier; what is more important is how you let others know. Many Silicon Valley founders spend 60% of their time every day promoting themselves on LinkedIn and X. Now there is a term called AI Wrapper—how to let your tool be seen may be more important than technology. So entrepreneurs who want to go global, you can even not write code first, put up a fake page to verify whether anyone registers. The logic has shifted from Product Market Fit to Market Product Fit—you have to first prove the Market exists, then make the product, don't silently finish it in three months then find someone to do SEO.

Ziwen: The most important thing in doing the Global market is to truly understand local users. Marketing is very important in the AI era. You can Build in Public, quickly polish a prototype on social platforms to see if users are willing to use it. Also, if you have theconditions, you can go deep into the local area—for example, if you want to hit the Japanese market, really go to Japan and see what knowledge workers are doing, attend local AI events and hackathons. Only by communicating with locals can you get real needs.

Xu Zuobiao: My view is similar to both of yours—Go to Market is indeed crucial. Sometimes when things don't work out there can be 100 reasons, but the core is actually that no one knows. When people ask you what your payment rate and retention rate are, it's all nonsense, because you have no traffic at all. If you have 100,000 visits every day, you can always keep some people to make adjustments for you—that is PMF. Some people feel that making a product doesn't need promotion and others will know about it—this is not very realistic. Finding the right channels to let others know and convert is crucial. Your real needs need to be polished and amplified; some products naturally start with low payment rates, or the early user profile is not clear.

Longyi: The main思路 is very similar. I add one point. When looking for whether anyone has demand, you can look for products that currently have enough traffic but are not AI Native—they may be further transformed. The demand must be large enough. If you盯 the global market and find only 100,000 people have demand, plus the conversion rate it will be very low. Also, regarding communicating with locals, we currently continuously communicate through WeChat and communities, as well as analyzing the Prompts users use in the backend. This can help us accumulate data for model post-training, while helping the team understand user scenarios, forming a flywheel between the team and data.

Shawn: Last question. Although it is only the first quarter of this year, I feel that every day there are new products disrupting old paradigms, making it hard to keep up. When future large model capability is amplified 10x, will your product still exist? Or what changes will there be?

Longyi: I think the model may be a brand-new OS that can run all data on it. For example, OpenClaw can help users串 together more tools in a very simple mode, and in the future will become a universal data conversion OS. Beyond data, what we are doing now related to "aesthetics" is not easy to be disrupted. If users input extremely simple prompts (such as minimal or cyberpunk), the content output by the model often tends toward "averageness." But humans need to see fresh things (new aesthetic Context), and we are actually doing this part.

Xu Zuobiao: If you are truly building product service scenarios based on large models, when the model gets 10x stronger, you would laugh awake from your dreams. Because you have high expectations for this scenario, and after the model gets stronger, those whimsical ideas or excessive demands you had two years ago can all be realized. You just need to solidly build the product and satisfy the scenario, so this is still a pretty happy thing.

Shawn: Time is about up. Thank you very much to the four guests for sharing. I hope your AI products go further and further in the process of globalization, and I hope the audience has gained something! Thank you all.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260420-01html
On-site reading page: https://ffcap.cn/en/research/src-20260420-01html