§ 1 The problem isn’t that you can’t code—it’s that you’ve been looking for the door in the wrong place

Over the past year, you’ve likely cycled through this pattern: You scroll past a post claiming "A college student built an app with AI in three days and earned $100,000/month." You click in, download the AI IDE recommended in the post, sign up for several model APIs, watch two or three YouTube tutorials—and then sit in front of a blank project in Cursor for half an hour—and close it.A few days later, you scroll past another one—and restart the cycle.

After enough time in this loop, one thought inevitably presses down:Am I excluded from this wave because I can’t code?。

That thought is wrong—but its error is deeply concealed.。

The real door isn’t on the side of "learning to code," nor on the side of "learning to use Cursor."It’s in another direction entirely.——Can you hear where real users get stuck—before anyone else does?Once AI tools slash the cost of turning an idea into a product, 99% the remaining 1% of scarce resources isinsight—not engineering.。

This isn’t an opinion. It’s something just validated at the Zhangjiang Science Hall in Shanghai,April 8–10, 2026.But before we turn to the hackathon, consider a cleaner case—a post by a post-2000 developer on Xiaohongshu: no product pitch, just a single question. Two weeks later, his team scrapped their original product plan. One month later, they launched an app that attracted 3,000–4,000 users in its first month—with zero paid promotion.。

His name is Sun Donglai. The app is Dreamoo.

When the post first appeared, his team’s product concept was VR dream interpretation—put on a headset and enter a visual reconstruction of your dreams. It sounded cool, and made for an easy fundraising story. That post was just a demand test: a single sentence—"Do you record your dreams?" No paid promotion; only organic platform distribution.Two hundred thousand to three hundred thousand peoplesaw it,and over 5,000 commentspoured in. One person wrote 800 characters daily, without fail. Another serialized her dreams on Tomato Novels for two years. A third left a long, detailed reply—so meticulous it resembled homework.

Sun Donglai and his team made a decision highly unusual for a startup:They scrapped the VR version and reverted to 2D.。

The rationale was right there in the comments. Most people who wanted to record dreams didn’t face a barrier of immersion—but of immediacy:Could they capture that fleeting image within 30 seconds of waking up?No one puts on a VR headset first thing in the morning—but people do open a 2D app. A month later, Dreamoo launched. In its first month, it acquired3,000 to 4,000 seed users (Tencent News originally reported 3,000; multiple outlets later cited 4,000—a distortion that recurs throughout this story), all from Xiaohongshu’s organic traffic.The app later entered the Chuhai Go Incubator (299 applicants, 1 selected—unanimously approved, verifiable on LinkedIn), and was featured at WAIC, the Amazon Web Services China Summit, and Y Combinator’s Demo Day. Some reports labeled it a multi-million-dollar AI startup—but no funding round or amount corroborates that figure. Treat it as secondhand commentary only. What *is* verifiable: one question, 5,000 comments, one pivot in product direction, and several thousand users in a month. At no point in that entire path did coding appear.

Sun Donglai’s path consisted of four actions—surveying, reading comments, scrapping the wrong direction, and relaunching—。

none of which are engineering tasks.Fig. ① Dreamoo’s path: Survey post → 5,000 comments → Dropped VR, switched to 2D → 3,000 seed users, zero code writtenCore insight

Dreamoo isn’t an outlier. It’s a phenomenon now scaling widely—but misnamed. Mainstream narratives call this "AI entrepreneurship." A more precise description is: The entry ticket to AI entrepreneurship is shifting out of engineers’ hands.

April 8–10, 2026, Zhangjiang Science Hall, Shanghai. Xiaohongshu’s inaugural Hackathon Finals, co-organized by Hillhouse Capital, with HarmonyOS as technical partner.

200 participants,

59 teams,48 hours,、closed development,、and a $500,000 prize pool.What’s worth examining is the shortlist of winners selected from over 60 submissions. Most coverage lists all six—but you don’t need to read them all. Compare three side-by-side, using Dreamoo’s path as a ruler—and you’ll see what this hackathon points toward.Grand Prize: Pocket Guitar,

team DAIZY, led by 22-year-old Ye Bowen, awarded$200,000.。

A guitar-shaped hardware device the thickness of a credit card, with real six-string fretboard, roller, and joystick. It played live during the demo—visually stunning, and rightly awarded. Butit wasn’t built from scratch in 48 hours— the team had prior hardware experience; those 48 hours merely polished an existing prototype into a demo-ready version. Its win reflected engineering skill plus aesthetic judgment—closer to a mature hardware hacker’s prototype showcase. Measured against Dreamoo’s path, none of the steps—surveying, reading comments, pivoting—have any counterpart here. Pocket Guitar followsan older path—technology-driven, visually arresting, but largely disconnected from user-painpoint-driven AI-era product development.Hardware Track First Prize: Brain-Controlled Wheelchair,team Mushroom Beef Noodle. Wang Ning, a wheelchair user himself and twice severely paralyzed, previously founded a startup using EMA (electrical muscle stimulation) devices. He claims he brought their cost down to under$30,000.

This time, he teamed up with his wife (a web novelist) and teammate Mira to build a hybrid control system combining EEG and head EMG signals. The headband design resembles a golden staff. During the live demo, Wang Ning moved the wheelchair forward, backward, and turned—using pure intention, hands off the control panel. Of the six projects, this is the one most worth remembering—real pain point + real end-user involvement in development + medical-device-grade scalability.Its narrative isn’t about Gen-Z technical showboating. It’s aboutthe person repeatedly failed by products taking matters into his own hands to fix them.Like Dreamoo, its demand was unearthed by someone who’d actually used the product—but unlike Dreamoo,Wang Ning didn’t need to post a survey thread—he *was* the comment section.The brain-computer interface SDK used wasn’t disclosed publicly. Don’t assume Neuralink or BrainCo.Software Track First Prize: Chic Chic,team When Haircut Meets Token Cap, building an AI hairstyle designer. Upload a photo; AI generates endless personalized hairstyle vibes, savable for TikTok hair transformation videos. Technically unremarkable—it’s image generation + face recognition + hairstyle prompt engineering—an AI wrapper application.

But itwrapped in the right place.Chic Chic follows Dreamoo’s exact path—first identifying where Xiaohongshu users get stuck (women, appearance economy, high cost of physical transformation—cutting hair wrong means three months to regrow)—then assembling AI tools to solve it. It proves one thing:You don’t need foundational innovation to win first prize—just deeper insight than others into where a specific user group gets stuck.Fig. ② Comparison of three winning project types: Hardware prototype showcase / Real-painpoint-driven / AI wrapper + user insightLaid side-by-side, Dreamoo’s ruler yields a direct conclusion—the hackathon wasn’t filtering for who can code best, but for who understands *where users get stuck.*Nearly all media latched onto the hook: "Youngest participant on-site: 12 years old." That number is true—but。

no report names him, states his school, describes his project, or quotes any judge’s evaluation.

It’s a widely repeated narrative symbol lacking first-hand detail—

don’t embellish it.Among youth teams with first-hand detail is Page One—four middle-schoolers (Jiang Muran, Lü Sitong, Yang Xizhe, Chen Yuxia), average age 13.5, who built Shuyi NoteRx in under 24 hours: a note-diagnosis tool for Xiaohongshu bloggers, winning the "AI Native Award." Technically simple, but hard in practice: these kids *are* Xiaohongshu natives. They know exactly what bloggers stress about daily—They don’t need to post a survey thread—they’re already standing in the comment section.Roughly tallying genuine innovation across the six projects: Only the brain-controlled wheelchair involved foundational technical breakthroughs. The other five were combinations of AI tools plus user-scenario insight. This isn’t a bug—it’s precisely what this hackathon aimed to say—Once AI tools are democratized, the scarcest resource in hackathon awards is no longer engineering ability—it’s insight. So among the six projects, who received the *real* entry ticket, and who got the prize money and spotlight? The distinction is clear:。

The brain-controlled wheelchair and Chic Chic got the ticket. Pocket Guitar got the prize money and camera time. These are two separate things.After reviewing all six, some will ask: Do the old rules still apply? Yes—they’re just being issued to a different cohort.§ 3 Where the Entry Ticket Has Shifted—from Where to Where。

What the old ticket looked like is well known to insiders: GitHub repos, open-source contributions, algorithm competitions, big-tech experience—accumulated over 3–5 years of coding, systems understanding, and patience. Winners fit a tight profile:STEM graduates, ex-big-tech employees, male-dominated, aged 24–35.。

That selection logic barely changed over the previous two decades—pre-AI.

The new ticket looks completely different. Spotting an undervalued real need on Xiaohongshu, assembling an MVP with AI tools, writing a post that algorithms want to distribute—these are entirely distinct competencies.The barrier isn’t coding—

it’s sensitivity to user context, patience to spend 1–2 hours daily scrolling and posting on Xiaohongshu, and psychological resilience to accept that nine out of ten products will fail.Winner profiles have also diversified. Over the past year, products emerging organically on Xiaohongshu include:。UI designers, liberal arts majors, high-school freshmen, HKU postdocs, French literature graduates, paralyzed individuals building their own medical devices, and web novelists who are stay-at-home mothers.The dispersion of this list—

would never occur via the GitHub path alone.But don’t interpret this as Xiaohongshu replacing GitHub.This is a fork—not a replacement—and distinguishing the two is critical—otherwise you’ll go down the wrong path.。

Some product categories remain inseparable from the GitHub path: infrastructure (databases, vector stores, inference engines), agent frameworks (LangChain, CrewAI), developer tools, and core ToB SaaS.LangChain could not be built on Xiaohongshu—its users debug in GitHub Issues, not Xiaohongshu comments.

Xiaohongshu’s path hosts other categories: ToC utility apps (Dreamoo, Weizhi Shu, Xiao Mao Bu Guang Deng, Honghong Simulator), lightweight hardware oriented around emotion and aesthetics (Good Luck Calendar Machine, Jike Fang P Mat), and micro-needs embedded in daily life (note diagnostics, hairstyle design).Dreamoo could not be built on GitHub—people who write 800-character dream logs daily aren’t on GitHub.

Fig. ③ Cross-platform comparison (core chart): GitHub / Product Hunt / Xiaohongshu / Jike—fork, not replacementThe old ticket (GitHub) and new ticket (Xiaohongshu) are issued totwo entirely different groups of people.

Both paths filter people—not rank them.

The required judgment is where your product idea sits—and then choosing the matching channel—not chasing whichever seems noisier.Xiaohongshu’s core metrics announced publicly at this hackathon were "350 million monthly active users, 160,000 AI developers, year-on-year growth, and 1.1 million Build-in-Public posts." Nearly all coverage quoted these figures verbatim.。But Xiaohongshu’s first public disclosure in April 2025 (re-reported by Tencent News in August) stated "50,000 independent developers, one-year growth of"—In one year, the number jumped from 50,000 to 160,000, and growth rate rose from 146% to 220%. Behind this, it’s unlikely actual growth tripled—more likely the

definition widened: from "independent developers" to "tech/AI content creators."Posting notes on how to use AI on Xiaohongshu is not the same as building products there as an independent developer—counting both under one metric makes the number look impressive.Key reminder: 160,000 ≠ 160,000 independent developers.220% Fig. ④ 2025: "50,000 + 146%" vs. 2026: "160,000 + 220%": Growth—or definition expansion?How many Dreamoos are in that 160,000? Likely zero were counted. Sun Donglai isn’t a "tech content creator," nor was he recognized for dev tutorials—he posted a product-survey thread and let the comment section redirect his product. People like him probably fall into the 160,000’s statistical blind spot—the very samples most representative of the new entry ticket are precisely *not* captured by that 160,000.

The assessment that Xiaohongshu’s AI startup ecosystem is expanding remains valid. But how many of those 160,000 are coding versus posting notes on how to use AI—Xiaohongshu hasn’t said, and no outlet quoting the number has asked.Next time someone cites "160,000" to urge you to "get on board," 146%the measuring stick is right here.The statistical definition has been broadened.——It expanded from "independent developers" to "tech / AI content creators."。Posting notes on Xiaohongshu about how to use AI is not the same thing as building products on Xiaohongshu as an independent developer.Lumping them together under one metric makes the numbers look impressive.

Figure ④: "50,000 + 146%" in 2025 vs. "160,000 + 220%" in 2026 — is this growth, or just a broader definition?

How many of those 160,000 are Dreamoo? Likely zero. Sun Donglai isn’t a "tech content creator," nor did he gain recognition through development notes — he posted a product research note and pivoted based on comments. People like him almost certainly fall outside the 160,000’s statistical blind spot —The samples most representative of the new opportunity aren’t propped up by those 160,000.。

The judgment that Xiaohongshu’s AI startup ecosystem is expanding remains valid. But how many of those 160,000 are writing code versus writing notes on how to use AI —Xiaohongshu hasn’t disclosed that breakdown, and no media outlet citing the figure has asked.Next time someone cites the 160,000 figure to urge you to "jump on board,"the yardstick for discernment lies right here.。

Of its 350 million monthly active users (MAUs), 160,000 are AI developers,0.05%The vast majority of Xiaohongshu users come for beauty, travel, and food content. The real status of the AI developer community on Xiaohongshu is:A small community living inside a high-traffic platform—an advantage (its traffic ceiling is extraordinarily high), but also a drawback (content must be written in consumer-facing, narrative language to be algorithmically distributed; a post titled “I Refactored the ReAct Framework Using LangGraph” will get no traction).

If you’re among the people this new entry ticket suits, the next question is: how do you actually begin?

§ 4 How to Do It: Four Actions—and One Counterpoint

The four actions, in sequence—each backed by Dreamoo’s own path.

Step one: Post a research thread first—don’t build a product yet.

The format is simple enough to write right now: “What’s the biggest pain point you face when doing X?” where X is your most familiar domain. A fitness content creator could ask, “Which step is hardest to explain clearly when recommending diet-tracking tools to your followers?” An independent teacher could ask, “What’s the most frustrating part about assigning homework to students?” A pet-product seller could ask, “When your cat is picky, what methods have you tried that didn’t work?”

Sun Donglai’s version was simply: “Do you remember your dreams?”—one sentence, no product pitch, no mention of “what I’m building,”just a question.The benefit of a question is that the comment section automatically reveals two things: whether demand exists at all (measured by engagement volume), and exactly where users get stuck (revealed in comments). Dreamoo’s 5,000 engagements signaled genuine demand;fewer than 500 likely indicates either fake demandor that the framing failed to reach the right audience. The goal here isn’t to land on an absolute number—but rather tosee demand intensity without writing a single line of code.The barrier is just a few minutes on Xiaohongshu—whether you can replicate itdepends on whether you’re willing to ask a real questionrather than polish a marketing slogan.

Step two: Treat the comment section as your PRD.

Don’t walk away after posting. Reply to every informative comment—and compile the most frequently cited pain points into a Top 5 list.That becomes your MVP’s feature list—no extras.。

Dreamoo dropped VR for 2D—not because Sun Donglai had a sudden epiphany one night, but because comments like “I never wear a headset in the morning” appeareddozens of times.He let the comment section redirect his product strategy. On this step,the more honest you are, the more you save.Every non-high-frequency feature you build is a cost you’ll later pay. This step is harder than Step One—it tests whether you’re willing to let users reshape your plan—many post research threads only to validate a direction they’ve already decided on, and instinctively ignore contradictory evidence.Sun Donglai’s outlier move wasn’t “listening to users”—it was tearing up the PowerPoint he’d already prepared after listening.。

Step three: Assemble your MVP using AI tools—don’t spend three months learning to code first.

Today’s usable tool stack looks roughly like this: an AI IDE (e.g., Cursor) for code generation; Claude or GPT APIs for core logic; Supabase for the backend; Vercel for deployment. Add a design-savvy co-founder or contractor,and you can ship a working first version in 4–6 weeks.For those who won’t write any code at all, try no-code tools like Lovable or v0 that generate full-stack apps from a single sentence—a demo ready for user feedback can be built in 1–2 weeks.Each tool takes just 2–5 hours to learn,far faster than enrolling in a three-month Python course.

Dreamoo had an atypical advantage here—Sun Donglai calls himself a “full-stack engineer”: a University of Science and Technology of China graduate student who can handle the entire tech stack himself. But in Dreamoo’s path,what truly mattered wasn’t this step,but the first two—if the direction is wrong, even a full-stack engineer can’t save the product.The real purpose of this step is to achieve “good enough”—don’t compete with professional developers on technical finesse.Outsource what you can’t handle—on Fiverr or domestic dev marketplaces,most early versions can be built for under ¥20,000.The most common trap is thinking, “I need to master the tech before I start.” That was the reflex of the last decade. In the AI-tool era,

starting immediately while learning on the fly delivers ROI on an entirely different scale—as you study, the AI tool stack itself updates monthly; knowledge acquired last year may no longer represent today’s optimal path.Step four: Launch cold using the same Xiaohongshu account.

Build in Public: Post 2–3 weekly updates on product development—not marketing copy, but process logs—e.g., “A user just told me… so I’m changing…” Screenshots of user feedback paired with iteration decisions carry natural algorithmic weight on Xiaohongshu because they signal

actual progress.real action—a strong signal to the algorithm.

Dreamoo executed this smoothly—its first-month user base of 3,000–4,000 came entirely from organic Xiaohongshu traffic,with zero paid promotion.But this is the least controllable of the four steps—the algorithm’s weighting mechanism remains opaque, and Sun Donglai’s posts gaining traction involved an element of luck. What you *can* control is amplifying that ‘real action’ signal—post ugly-but-functional versions,screenshots of user criticism, or your internal debate over cutting a feature.Don’t wait for “perfection before posting.”Early users tolerate far lower completion thresholds than you imagine—as long as you solve a problem they genuinely care about.

Together, these four steps aren’t a product-building manual—they clarify what thenew entry ticket looks like—it’s not a certificate,but a set of concrete actions.，Each step can be practiced starting tonight.。

The counterpoint must also be stated plainly—who should *not* take this path?

Those building infrastructure, agent frameworks, or developer tools should go to GitHub, V2EX, or Jike—users aren’t on Xiaohongshu.

Those building B2B SaaS—Xiaohongshu has extremely low density of B2B decision-makers—seek vertical-industry sales channels instead.

Those unwilling to spend 1–2 hours daily browsing Xiaohongshu and posting updates—this path isn’t for you. It’s consumer-facing, content-driven entrepreneurship—not a single-point breakthrough.

Those unable to accept that “your first 10 products will likely be garbage”—Chen Yunfei and Wang Dengke have both stated outright:independent developers typically need to ship 10 products before one breaks through.。

Those unwilling to “lower their stance” to build for students, humanities majors, or older adults—Xiaohongshu’s real traffic lives in these groups, not in any professional mutual-help circle.

Fig. ⑤ Four-step workflow: Research Post → Comment Section → MVP → Build in Public (each step annotated with cost and time)

Completing these four steps sounds straightforward—butthe hardest part isn’t the steps themselves.。

§ 5 Closing: This path is 100× cheaper—but not free.

The hardest part is accepting thatthe first nine attempts will fail.

These are authentic voices from within the industry—not meant to scare you, but to show you the true numerator-to-denominator ratio.Yet this same cohort serves as the most credible witnesses to this path’s viability.Because the cost of a single failure has changed.。

In the past, failing on one product meant losing several months plus hundreds of thousands in startup capital—depleting personal savings was common.Today, failing on one product might cost just a weekend plus a few hundred dollars in API fees. Failure cost has droppedby two to three orders of magnitude,meaning you can afford dozens of times more attempts.Success probability hasn’t increased—but total time to reach success has shrunk dramatically—that’s the most valuable outcome of AI-tool democratization—not any single model getting stronger.

The biggest shift in this generation of AI startups isn’t that models got stronger—it’s that the cost between “having an idea” and “getting users” has been slashed.The entry ticket still exists— 99% it’s just moved from GitHub to an app everyone already uses.Back to Dreamoo’s origin: At the moment Sun Donglai posted “Do you remember your dreams?”,。

he had no idea it would gain traction.No paid promotion plan, no KPIs, no waiting until every feature was fully designed—he simply dropped the real question in his head onto Xiaohongshu as-is,and let the comment section answer it. This action requires zero technical skill—the barrier is just a few minutes and hitting ‘post’.Whether you earn this entry credential tonight depends not on coding ability—but on willingness to repeat that same action:。

Open Xiaohongshu, search your most familiar domain plus ‘AI’, and see what others have built—and where they’re stuck—at that intersection. Tomorrow, carve out time to post your own:。

Turn the problem you’ve observed in your field—one that AI hasn’t solved yet but should—into a single question.Engagement numbers will give you the answer.。

Views expressed in this article are for informational purposes only and do not constitute entrepreneurial or investment advice. Sources: Tencent News / GeekPark / IT Home / Zhidx / Guancha.cn / LinkedIn / Xiaohongshu Hackathon official announcements.。