FOMO won't take us in the right direction; the goal is to stay in the game

AI Applications | Negative Sentiment | Truth | FOMO

Intro

Welcome to China's AI startup scene in 2026 — a world where noise and truth are wildly out of sync. A year ago, AI applications were the darlings of the primary market. In 2026, investors say they've stopped looking at AI applications altogether.

"Those of us building AI applications became, overnight, like old toys tossed in a corner — no one wants us anymore," says Zijian Jia, CEO of Inspired AI.

Jia won the AI Night Hunt in Shanghai. He entered the AI industry in 2023 and has two language education products under his belt: TalkMe (AI-powered spoken language practice) and ListenLeap (podcast-based English learning). The former has traded the top spot with Duolingo in Taiwan; the latter has thousands of organic user posts on Xiaohongshu. Together, the two products generate several hundred thousand dollars in monthly revenue and turned cash-flow positive early on.

During the Mobile Internet era, he was also an accomplished product manager, accumulating ten years of consumer product experience at NetEase, 360, and TAL Education Group. He built dozens of products with millions of DAU and two products with over 100 million daily active users.

In August, he launched a third product — an information provenance tool.

He wants to set the record straight for AI application practitioners. His honest take: the idea that models will swallow most applications is absurd — it assumes technological evolution automatically equals commercial substitution. "Spring for AI applications is far from over. The relationship between applications and large models isn't black-and-white; it's an ecological, systemic integration, an upstream-downstream interdependence, and mutual complementarity in vertical domains."

Below is Jia's share. His perspective deserves attention because he represents a silenced voice in the 2026 AI application narrative — a project that doesn't chase hype, doesn't fabricate narratives, doesn't subsidize users with investor money, yet achieves stable profitability with solid margins. In a negative funding climate, how does one survive and adapt without being swayed by external forces?

"Regardless of the funding environment, we entrepreneurs must always answer one question: Who are the users? Why do they pay? Can we make money?"

Black and White

"We didn't invest in them for this. They've pivoted three times this year." That's what an investor friend of mine said recently when we were discussing a certain product.

I couldn't tell if he was complaining or just describing things objectively. The truth is, all entrepreneurs fear making mistakes — especially when you spend half a year building something, only to wake up one morning and find it rendered obsolete by a model upgrade. So you pivot again and again, try this and that.

This investor also told me that this year is a big year for investing. Embodied AI, world models, compute, quantum computing, controllable nuclear fusion, brain-computer interfaces... a spectacular blooming scene. Whether people are genuinely paying attention or actually writing checks, everyone's busy and overwhelmed. "But AI applications? We really don't want to waste any more time on them."

In less than a year, those of us building AI applications and agents became like discarded old toys, thrown in a corner, ignored.

The root cause, according to many industry insiders, is the belief that AI applications will eventually be swallowed by large models.

But I think this is FOMO, and more than that, a lazy question.

By lazy, I mean most people think at the surface level, treating the relationship between AI applications and large models as black-and-white, when they are originally an ecological, systemic integration — upstream-downstream interdependence and mutual complementarity in vertical domains.

Having built products for over a decade, and having started my AI entrepreneurship journey in 2024, I have a basic taxonomy of products that I'd like to share, to help explain why I think this view is lazy.

Whether AI is involved or not, products fall into three categories.

Category one: outcome-delivery products. As foundation models get stronger, some products do trend toward merging with large models. But products that build unique datasets and proprietary models in vertical tracks, creating significant differentiation in outcome delivery, are extremely valuable. I know entrepreneurs working on specialized medical and legal documents, for instance. There's also a Hangzhou company called Xingse Shihua (Plant Identifier) that makes over $100 million USD annually. Large models identify plants and flowers with 60% accuracy; they hit 95%.

Even Manus initially focused on optimizing depth for individual vertical scenarios and niche user segments, building data and engineering moats.

Category two: content consumption products. Douyin is the prime example. Humans will always need to scroll through this kind of product because dopamine demands it — only the product form changes. Didn't Douyin recently launch an AI Douyin? There are plenty of opportunities here too.

Category three: process-delivery products. Games are the standout example. Then there's our own track, AI+Learning. As models evolve, products in this category will inevitably become smarter, knowing you better than you know yourself, with ever-improving user experience. That's where we sit.

TalkMe delivers "the process of opening your mouth and speaking." Many people study foreign languages for years but still can't speak fluently, because no one practices with them. This product is always there to practice with you. ListenLeap follows the same logic: it turns content you love into listening material, making you more motivated to actively learn English.

As models get stronger, this category of product will know your weak points, know what content you like, and user experience will only get better.

Based on this taxonomy, those who say "AI applications will be swallowed by large models" are indeed being lazy.

Four Strange Things

In the first half of this year, among the investors I've met, I've genuinely felt capital's negative sentiment toward AI applications. At the same time, I've witnessed many strange things.

The first strange thing: the amplification of AI product buzz.

Most products' communication channels are Twitter and Jike. They look massively popular, like they blew up overnight. But I've never quite understood — aren't the people reposting on these platforms just programmers and AI practitioners? Isn't this still just the industry circle getting high on itself?

Several of these products I've tried, and I couldn't quite figure them out. I don't know who would pay for them.

I've heard even more outrageous stories. Early this year, an interactive product got liked by a big-name figure, and domestic praise poured in. Later I heard that big-name likes can be bought now — there are specialized companies for this. The founder spent hundreds of thousands buying likes to keep raising funding.

The second strange thing: an investor recently told me, "Zijian, I don't think you're AI-native enough. Your team isn't AI-native enough either."

I asked, "What is AI-native?"

They couldn't answer either. "Anyway, it just feels like your team isn't AI-native enough. Too old-school."

Later I figured it out. What they call "AI-native" probably means something particularly fancy — flashy interfaces, animations and effects everywhere. Ideally the founder is a post-00s wunderkind, back from studying abroad.

Or throwing around big obscure terms like "General Agent + Skill + MCP." These are industry jargon. Ordinary users don't care. They care about one thing: can you solve my problem?

Both my products target the most ordinary users imaginable. If they can't understand the product, they're gone in three seconds.

So I don't quite get it: are some products built for VCs, or for users?

The third strange thing: investor sentiment changes faster than the weather.

Many investors will instruct me: Zijian, you should go in this direction now. A few days later: If you pivot that way, it'll be easier for us to get you through our investment committee.

Some young founders have told me they feel lost because investors have a new direction every day.

I understand that sometimes founders genuinely can't find direction and are extremely anxious. To secure investment opportunities, they really do take investors' words to heart.

But every pivot costs money, people, energy, time... The cost is high. What if it fails?

Investors never take responsibility for their guidance. Founders are the first people accountable for their products. Sometimes a ship keeps changing course because the captain hasn't thought clearly enough.

The fourth strange thing: things that are genuinely useful to users lack attention.

It seems like the market no longer cares about real user needs. I often joke that many real needs aren't on X — they're at the wet market.

My aunt, 70 years old, uses Doubao every day to generate text and images. She also does live commerce on Douyin, making 7,000-8,000 RMB a month. She told me her scripts are basically all AI-generated. I was shocked.

There are many products like this — ignored by the primary market because they're not fancy enough, not buzzworthy enough, but solidly built, with scenarios, users, and viable business paths.

To use our own products as another example: I don't understand why the industry thinks language learning is so simple. Maybe because AI practitioners are just too good at learning?

But for most ordinary people on Earth, learning a language is excruciating. It requires a teacher's personal instruction, someone to practice with long-term, someone to push you to keep going.

Facing the Brutal Reality

The primary market is this strange right now. We still need to survive, still need to keep moving forward. So what do we do? Or put another way: What determines victory or defeat for AI applications?

My answer: Products are always built for users.

Let me share my experience from building two products with over 100 million DAU.

First: Opportunity hides in user reviews.

Many technically brilliant people want to start companies but feel there are no opportunities left, that the big players have taken everything. I disagree. The opportunities in AI applications are far from fully mined.

My initial method for finding product-market fit — you might not believe it — was that I decided early on to focus on education scenarios. Then I went to the app store, found the best-performing products in the same track, and read every single review.

Thousands of reviews, one by one. Don't rely on AI summaries — total nonsense. You have to read them yourself, feel them yourself.

Then go to Xiaohongshu and see what real users are discussing. Look for grounded, specific comments, especially criticisms. After reading these, you'll definitely find opportunity points.

To this day, the first thing I do every morning at work is read user reviews. Many founders stopped paying attention to these details long ago.

Whether you're carbon-based (human) or silicon-based (agent), the core path is: where are your users? Don't deviate from this.

Second: Control team size.

After finding PMF, start with 3-4 people to test.

When TalkMe became number one in Taiwan, the team had only 7 people.

Never expand the team too quickly from the start. Because funding is hard to come by, I often tell myself: this round of funding is probably the last I'll ever raise in my life, so I must spend it carefully.

This isn't conservatism. It's clarity.

For an early-stage team, ten people maximum. No more.

Third: Build a product that can be explained in one sentence, and go deep on operations.

If you can't explain it in one sentence, users are gone in three seconds. Users won't sit through your long explanation.

We chose a scenario with stable demand and supply on both sides. The upside: the scenario will exist long-term. The downside: it's competitive. So core capability becomes product operations capability. You may only have 1,000 users now, but retain them, and you'll slowly reach 10,000.

As a product person, I've built dozens of products with millions of DAU and two with over 100 million daily active users. In my humble opinion, sustained, long-term operations are the true test of a founding team's core capability, because most companies struggle to survive the 3-5 year cycle.

Go deep, do well what you can do now, and accumulate models, data, and operational experience — these are the long-term determinants of victory. Vibe coding can copy everything visible, but it can't copy your ability to acquire 1,000 → 10,000 paying users.

Fourth: Do the math. Don't lie to yourself.

I have an iron rule — within two months, if user payments can't cover customer acquisition and token costs, pivot immediately, because the product has no demand.

Yes, 60 days to test PMF.

This sounds brutal, but it's the reality of AI applications. Because all products consume tokens. Today's startup environment doesn't afford the luxury of slowly nurturing users.

I've talked to many founders whose products have decent revenue. The feeling is the same: sometimes within a week of launch, based on user payment behavior, experienced founders can roughly judge whether the product has real paid demand.

Of course, there are also many products that maintain high growth, mostly by burning money to subsidize users. I don't understand how they make money — there can't be profit.

Later I figured it out: they're using investor money to subsidize users. The investors in these projects probably aren't operating on pure math logic either. Maybe there's always a greater fool.

Finally, if your product happens to land on a big company's extension line, don't think about fundraising. Small and beautiful is fine. Don't deceive yourself, expand the team, raise funding, inflate investor expectations, and end up miserable yourself.

Learning to quit is also important strategic thinking.

Survive First

Fundraising is genuinely cold this year. Some investors say, Zijian, you missed the window. Why didn't you raise last year?

At first, I felt some regret. But on second thought, it might not be a bad thing.

I now have two products with stable monthly income. So why not rest — not rest, recalibrate — and as a result, I came up with a third product this year.

One-sentence intro for the new product: It assists users in making high-quality decisions, using exclusively firsthand information.

This addresses my own pain point. As a CEO, I make massive numbers of decisions daily. I want to see firsthand information.

You ask a question; the product doesn't give you an answer directly — instead, it presents various viewpoints, data chains, evidence chains, and papers as a structured briefing, including which concepts each viewpoint references. All original statements, original viewpoints, original context.

Unlike large models, we don't index web pages, because many web pages are also secondhand information. Large models excel at "generation"; we focus on "provenance."

In early testing, several user types have already clearly emerged. First, people doing research. Second, people making various business decisions daily — such as primary and secondary market practitioners. Third, debate teams — a scenario I didn't expect.

The product direction benchmarks against Bloomberg and AlphaSense.

Actually, for a first-time entrepreneur, choosing this direction has quite a high barrier. But on the foundation of having successfully built two language learning products, we possess mature experience and accumulated沉淀 in information provenance, models, data, content, and algorithms.

Another advantage: the new product forms a product matrix with the previous two, sharing traffic and users, with similar operational playbooks.

So smooth fundraising isn't necessarily a good thing. Lying low and biding time might lead to new directions.

As large models evolve, products will exist in phases in this world. Entrepreneurship is inherently a marathon. First ensure you're still at the table, then think about reaching for the stars. I've talked with Manus too — they also survived several cycles.

A few days ago, an old shareholder came to chat. He encouraged me: Zijian, the cycle is indeed tough right now, but if you can survive it, your team will level up.

This comforted me. How does that saying go — many great companies in history were born in bear markets.

This is also the best time for truly capable entrepreneurs and investors to find suitable opportunities. I've heard some fast-moving investors have already started looking back at applications.

FOMO won't lead us in the right direction. It will only disrupt our footing. If a ship keeps sailing in one direction, it's because the captain is sufficiently resolute.

Cover image provided by | Zijian Jia

This was also his pitch deck cover

AI Night Hunt is looking for independent contributors

If you're curious about the people and stories in the AI wave, and skilled at writing

Please contact: shizao1123@hotmail.com

Email should include: 1. Personal introduction; 2. Relevant articles; 3. Your favorite AI Night Hunt article

Competitive compensation, professional editorial guidance

Looking forward to making noise with you