AI Won't Make Humans Obsolete in a Decade: Chen Mian's All-In Wager
If AI can't learn human taste and judgment within a decade, design products like Lovart that require human curation stay valuable; bet wrong, and the company fails with it.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Seven job changes weren't drifting; they were filtering
In a decade, Chen Mian cycled through seven or eight roles across nearly every major Chinese tech company—Tencent, 360, Baidu, Didi, Mobike, Daily Fresh, ByteDance. He calls it not wandering but deliberate filtering: whenever he judged he couldn't become a decision-maker and driver at a company, or if there was nothing left to learn, he'd move. When recruiting for his startup later, one prospect hesitated to join because he'd ‘jumped too many times’. His answer: if it's ultimately about starting a company, past job-hops don't matter—what matters is the decade of samples, which let him read almost every major Chinese internet battle, predict the winner, and explain why.
— Chen MianMobike taught him: product isn't the winning lever
When Chen Mian ran bike-finding and repair at Mobike, he discovered business outcomes had nothing to do with product experience—users only needed to scan and unlock, not even the app. The real deciding factor was vehicle maintenance, repair costs, and whether operations could break even. This shattered his self-image as a product manager: at both Didi and Mobike, product wasn't the deciding force; operations and supply-chain were. This insight shaped how he judged what matters when starting his own company.
— Chen MianSubscriptions are actually a win-win for all sides
When Chen Mian commercialized Capcut from an internal Douyin tool into an independent product at ByteDance, the shift from one-time buy to subscription was key. He kept adding templates and materials (text effects, stickers) as justification for ongoing payment. He sees subscriptions as ‘clearly a win-win’ for both business and user—users avoid the sticker shock of buying Office for five to six thousand yuan upfront. But he notes Chinese consumers' willingness to pay for digital goods remains weak, a long-term handicap for China's software industry that government support can't fix.
— Chen MianQuit before the plan: move on inflection instinct
By late 2022, Stable Diffusion suddenly ‘got so good’ and ChatGPT arrived in succession—two shocks that convinced Chen Mian this was at least a ‘computer invention’-scale revolution, no smaller than mobile internet. He judged the window for image and agent products as only a year or two before 2024—he had to move before the explosion. So in May he quit cold: no fundraising talks, no co-founder lined up, no heads-up to colleagues. Just ‘the way to hustle is to hustle first’. He even forewent the major equity payout waiting in his final year at ByteDance.
— Chen MianFour thousand yuan left—he covered the gap himself
After LibLib launched, Chen Mian initially said yes to every VC that approached via ‘equal rain on all fields’—treating each investor the same—which led to low valuations and fewer options for subsequent rounds. Then a deeper-pocketed competitor ‘burned cash straight at them’ to undercut him. At the crunch, the company had four thousand yuan left—he put in his own money to carry it through the final two months until cash flow turned positive. Meanwhile, he was laying people off while managing team speculation about ‘whether they'd be acquired’ and morale collapse. His conclusion: ‘doing the business well doesn't mean you'll raise money’—a belief he now knows was wrong.
— Chen MianDon't compete downstream of the megacorp; go upstream
Chen Mian has boiled down how AI startups escape megacorp squeeze to two rules. First: don't sit downstream in what megacorps do (photo retouching will never beat Photoshop and Meitu Xiuxiu)—be upstream. Midjourney and LibLib's users flow downstream to Photoshop, sure, but that hasn't stopped them scaling. Second: serve new users or new needs megacorps can't reach. He cites Canva relative to Adobe as proof. By this frame, Lovart isn't up-downstream from LibLib; it's new need entirely—the goal is to let ‘clients who'd hire a designer’ do it themselves. Essentially: replace part of the designer role, not just offer a better tool.
— Chen MianChinese AI winners are battle-hardened, not hungry newcomers
Chen Mian observes that most winning Chinese AI startups are run by people who were chomping at the bit in mobile internet days—battle-worn founders. Young newcomers are rare. He thinks this differs from Silicon Valley, where founders might breakthrough on single insight, ship a great product, and sell it—exits are smooth. But China's payment culture and competitive intensity demand founders who can do business, fundraising, and growth all at once. Young people rarely break through on single-strike ability alone; it's the prior generation—the ones who fought hard and hadn't yet cashed out—who can catch this wave.
— Chen MianAGI won't beat humans within a decade—Chen's bet
Chen Mian openly says he doesn't believe AI will fully surpass humans in five to ten years—it's not optimism, it's Lovart's product assumption bedrock. If he truly believed AGI was imminent, there'd be no point building anything but the model itself. But if human taste and judgment stay relevant five to ten more years, Lovart should let AI models that are ‘nearly human-level’ be used the way humans talk to humans—which is why they built an interactive canvas, not just a chat interface. He admits: if AI truly surpasses humans across the board, it'll surpass by ‘a mile’, and the company dies with it.
— Chen MianIn their own words · checked verbatim
I'm someone who really wants to win. And I think I really want to—really want to do something myself, really want to make my own decisions, change something. Yeah. That's my driving force.
我是一个很想赢的人 然后我觉得我是一个很想 很想自己做点什么 很想自己决策 改变点什么的人 对 这个是我的原动力
Chen Mian39:20
You discover that only this one thing matters, everything else doesn't matter—everything else is just subsidy.
你就发现只有这个东西是重要 其他东西都不重要 其他东西就是补贴重要
Chen Mian53:29
By the end of '22, suddenly—how did this get so good? Yeah, it felt like science fiction. Yeah, that's when I got completely blown away.
到22年底的时候突然 怎么这么好了 对 感觉像科幻了 对那个时候我是被 被Shark到的
Chen Mian1:30:50
I've seen an exaggerated report saying we ended up with only four thousand yuan left in the account—literally, on the last day, we had exactly four thousand yuan.
我看过一个很夸张的报道 说最后账上剩四千块钱 最后一天就是只有四千块钱
Chen Mian1:55:11
One: you shouldn't position yourself downstream of someone else's workflow, but you can position yourself upstream. That's a particularly good point.
一个是你不要做在人家工作流的下游 但你可以做在人家工作流的上游 这是一个特别好的point
Chen Mian2:20:25
I don't think it's possible. I don't think it'll happen within a decade. I think within a decade it's very hard.
我不可能 我觉得十年内都不会 我觉得十年内都很难
Chen Mian3:04:49
He felt that he helped tens of thousands of people—people unhappy with their appearance, people with face anxiety—make themselves look better. And that made him happy.
他觉得他帮助了成千上万的人 对自己长相不满意 对自己有面容焦虑的人 把自己弄好看 然后他心情很愉快
Chen Mian3:18:00
Because it's just—make a scene, then slip away quietly. That's my wish too.
因为就是 就叫大闹一场 然后悄然离去 这也是我的愿望
Chen Mian3:21:02
Figures
| LibLib account balance at tightest moment | Four thousand yuan | 1:55:11 |
| Guagualong (瓜瓜龙) product team size at ByteDance | Over one hundred people | 1:02:32 |
| Lovart's first promotional tweet impressions | Over one million | 2:57:45 |
| Lovart current team size | Fifty to sixty people | 2:25:27 |
| Office price Chen Mian cited in early startup days | Five to six thousand yuan | 1:14:36 |
| Chen Mian's rank at ByteDance | Level 4-1 (among the company's youngest ever at this level) | 1:16:38 |
Glossary
- AI native
- A product whose value entirely depends on AI capability—remove the AI and the product fails. Not an older product with AI tacked on.
- reward model
- In AI training, the feedback signal that judges correctness. Chen Mian borrows it metaphorically to describe the psychological mechanism for validating your own competence.
- AGI
- Artificial General Intelligence: AI that surpasses humans across nearly all cognitive tasks.
- Agent
- An AI program that breaks down objectives independently, then executes multiple sequential steps without human guidance at each step.
How to listen
Founders, product managers, and investors who need to understand how AI startups dodge megacorp dominance and spot their window.
The first 20 minutes of childhood and family background chat—doesn't help you understand the startup logic that follows.