"Why Hasn't the AI Super App Emerged Yet?" | Ten Reflections, a Glimmer of Hope
Dreams are worth having — what if they actually come true?
🚦Author: Orange, former Hailuo AI product manager at MiniMax, expert in AI productivity tools and content community products.
1. Prologue
The AI industry lately seems to have fallen into a collective anxiety. Many friends have come to me, puzzled, to talk about AI Super Apps:
Why haven't we seen a true AI Super App yet?
AI technology looks impressive, but the numbers for AI products are mediocre. Where exactly is the problem?
The business models for AI products are equally baffling. Most use subscription pricing now — so why are conversion rates so low?
I've thought about these questions for a long time. The deeper I went, the clearer one thing became:
An AI Super App can't rely on AI alone. It must mobilize people — building a system where everyone can participate.
2. How Do We Define an AI Super App?
There's no strict definition of an AI Super App, and I'm not going to dwell on definitions.
But since it's a Super App — a future app for the masses — it needs at least tens of millions in daily active users.
Currently, only one AI-native product hits that mark: ChatGPT.
But it's an outlier. Last year, it exploded through a combination of technology that was a year ahead of the industry and a marketing-genius CEO, seizing mindshare through breakneck growth.
It had perfect timing, positioning, and execution — unrepeatable.
Look at everything else: 30-day retention for new users doesn't even hit 20%, and DAU struggles to break ten million.
By retention metrics, Character.AI is the exception — its 30-day retention cleared 20%, and since it's in entertainment, it reached 7 million DAU without heavy paid acquisition. No other AI-native product has cracked 3 million DAU.
To put that in perspective, in the mobile internet era, a few million DAU was nothing special.
Neihan Duanzi — nobody would call it a Super App — had 20 million DAU.
BOSS Zhipin, a job-search app, had 15 million DAU.
And now Doubao + Kimi + Wenxin + the entire field combined don't even reach 10 million DAU.
That's after burning cash down to 30 RMB per CPA.
10 million out of 1.4 billion — a penetration rate of 0.7%. Pathetic.
AI claims to be a technological revolution. Why is the situation so grim?
3. Why Is AI Product Penetration So Miserably Low?
Let's think about this from first principles.
The essence of a product is satisfying user needs.
You can brute-force new user numbers with cash burn and paid traffic, but usage frequency, time spent, and retention don't lie — users vote with their feet.
At its core, AI isn't meeting mass-market user needs.
Or put another way: it's not solving real problems for ordinary people.
Users want to find jobs; AI can only help polish their resumes.
Users want to find partners; AI can only draft chat replies.
Users want to buy things; AI can only offer some selection advice.
What users actually need is matching, brokering, action.
4. What's the Biggest Difference Between AI and Humans?
Let's go deeper, to the essence of human exchange.
Why do humans need to connect and transact with each other?
Let's crudely break down human value into two main things:
- Providing value that helps you make money
- Providing emotional value that makes you happy
Can AI help you make money?
Yes — if you use AI to create a product or service, then sell that to customers. So on the B2B side, AI really can help you make money, and indeed B2B AI data looks strong.
But on the B2C side, let's not even talk about ordinary users — take creators. Have GPTs and Coze creators actually made money? No.
Can AI make you happy?
Yes — chatting with virtual characters can bring some fantasy-tinged pleasure.
But compared to short video, games, and social apps, that happiness is pretty thin.
Whether ad revenue or paid revenue, it can't match traditional entertainment without AI.
5. Solve Problems, Make Money
The more common way humans solve problems isn't by themselves — it's through other humans.
We build software by hiring product managers and engineers.
We get rides, food delivery, and moving help by paying people to solve our problems.
Pay money, get the result you need.
The entire human economic system is built on this simple model.
One person helps another solve a problem, and they earn money. AI too needs to truly solve user problems before it can earn money.
Yet AI currently has two inherent problems:
-
Capability constraints: with current AI, paying money doesn't guarantee a solution. It's a probabilistic product, full of hallucinations, to the point where it has to use the internet as RAG and become an "AI search engine" just to function as a knowledge retrieval engine.
-
Even with sufficient capability, AI alone isn't enough — human collaboration and coordination are needed. Humans accomplish things not through sheer intelligence alone, but through help and cooperation from others. Humans are symbiotic. Each person accumulates unique knowledge and skills through long growth, and the complementarity of different people's knowledge and skills enables them to accomplish bigger things together.
A CEO's greatest responsibility isn't to be personally brilliant — it's to find capable people and build the team.
So some think: what if we had Agents, each excelling in their domain...
Couldn't a CEO just assemble a team of Agents to get big things done?
Unfortunately, we're discussing technology, not science fiction.
6. Agent Is Still Science Fiction
There's a boundary between science fiction and technology. If the full score is 100, you need at least 60 to count as technology.
Optimistically, Agent is currently below 20. So for now, Agent is basically a science fiction concept.
The most sci-fi branch is prompt-engineering Agents.
"You are a coding expert, proficient in all programming languages. Your code is logically clear, highly practical, and bug-free. Now, please write a Douyin app."
You might find this prompt ridiculous, but in many AI products, this is literally how users write prompts.
This isn't just sci-fi — this is magic!
So counting on UGC to build an Agent platform is merely a naive, beautiful wish.
7. Back to the Essence of Making Money — Is AI Solving Problems?
AI needs to truly solve user problems before it can make money.
The current consensus is that AI is just a Copilot — its own capabilities are insufficient and need human guidance.
Given this technical consensus, ultimately solving problems and making money still depends on people.
The people directing AI are developers.
So has AI enabled developers to make money?
At least from B2B observations, yes. Overall B2B is in a state of supply shortage — enterprise demand for cost reduction and efficiency improvement is widespread and urgent, but developers who can meet enterprise needs are far too few.
Previously, developing enterprise intelligent services required a fairly large R&D team. Now with workflow scaffolding like Dify and HiAgent (Coze's enterprise version), building AI products takes just two or three people — the barrier to B2B service has been lowered considerably.
But lowering the barrier to small teams is still a barrier.
Many individual developers are quite smart and can equally satisfy various enterprise needs, but they lack B2B business experience and don't understand B2B playbooks. And on the B2C side, they also can't directly make money by building Dify or Coze products.
Imagine if high-quality content on Douyin required teams to produce — Douyin wouldn't have the prosperity it has today.
This is the industry's current problem: the barrier to AI app production still isn't low enough. It needs to drop to the individual level.
Only by solving this industry problem can everyone make money.
8. AI for Everyone Isn't Just a Slogan
Everyone's shouting "AI for Everyone," but how can that happen without solving industry problems?
The core reason individual developers can't make money building AI tools is the lack of a complete platform.
Some say: don't we have Coze and Dify? Isn't that enough?
No, it's not enough, for two reasons:
-
Dify is a developer-facing product. It's weak on the publishing side — hard to publish a product-grade page or mini-program, let alone monetization.
-
Coze can publish to many places, but its business logic is SaaS logic. It can't integrate other models, leading to high costs, and it hasn't built money-making features for users. So creators build things, hand them to users, and users have to pay Coze instead — the business model isn't platform win-win thinking.
A ray of light
However, Wordware AI, which went viral recently, solved both products' problems while also doing one more important thing, assembling three key elements:
- Can publish as a website
- Can charge directly on the website
- Can copy website templates for rapid site building
These three simple things essentially build an AI site rapid-deployment system.
Advanced players can develop websites themselves, no platform needed.
Intermediate players can use Dify or Wordware to create original workflows.
Beginner players can copy successful website templates on Wordware and monetize through their own growth tactics.
The largest user base is beginner players. Wordware lowered the site-building barrier to every individual.
You could call them AI-era webmasters, or AI product solopreneurs.
9. If Everyone Can Be an AI Webmaster, Could the Platform Become a Super App?
The prosperity of search engines was inseparable from PC webmasters. After ICP filing policies, PC websites declined.
The prosperity of e-commerce platforms was inseparable from small merchants. When Taobao abandoned some merchants, Pinduoduo took them in.
The prosperity of food delivery platforms requires not just merchants but also riders. Upstream sparked huge controversy precisely because of public attention to the rider community.
So could AI tools become an AI tool platform?
I think it's possible.
Except last year's hype had AI companies fantasizing about achieving this through Agents — that's a sci-fi story.
While many tools have introduced UGC, the content industry has always been hits-driven; the essence of UGC is actually PUGC.
One great piece of content is worth more than ten thousand mediocre pieces.
One great tool is worth more than ten thousand mediocre tools.
Since AI is still a Copilot, the ones solving problems in the platform are still the people directing AI.
With people directing AI involved, they need to be able to make money.
Let professional tool makers make money.
Let beginner tool makers make money through COPYING.
Let traffic-holding tool distributors also make money.
Only then can the platform make money.
Build a system where every participant wins.
Right now, this may be the pragmatic path to creating a Super App.
10. Conclusion: Respond to Change Through Constancy
The above is thinking based on the current state where OpenAI can't produce a breakthrough and the whole industry is somewhat stalled.
Maybe after GPT-5 launches, everything changes, and Agents actually work.
That's a variable — unpredictable.
But one thing will certainly not change:
Human nature chases profit. To attract people, everyone must see a path to profit.
