Original · Unique Research · 2026-06-24
Editor's note: The first-person report and its judgments belong to the original Chinese author (hosted by Wu Wei; interview subject Bai Ya). This English rendition retains the narrative on the Songti tell, experience as a liability, role merging, the AI-transformation bottleneck, customer-service economics, personal advice, and coda, plus the complete 20-question Q&A. All named people, companies, and figures are preserved. Founder statements and cited numbers are source attributions, not independently verified findings.
AI Industry Observation
Eliminate yourself first; rest when it's time to rest
The real starting point of the AI era isn't the one-person company, but actively replacing your old self
"
The one-person company isn't the endpoint. Eliminating yourself first is the starting point.
Why Do the Chinese Titles on AI-Generated Web Pages Always Use Songti?
Bai Ya threw out this question on a livestream. He's a product person trained in fine art who spent over two decades in design, founder of Youzan and superun.ai.
The answer is a bit interesting. In the early days browsers rendered poorly; Songti on a screen was all sharp corners and jagged edges, so body text on Chinese web pages must never use Songti — an iron rule. But graphic design follows a different logic — in books and posters, traditionally toned designs often use Songti for titles: elegant, breathable; slam a heavy heihei onto a big poster and the image gets crushed. This knowledge isn't written in any programmer's spec book. When models trained, they devoured countless programming textbooks but never ate "don't use Songti for Chinese web titles." AI scientists don't think aesthetics is worth their time. As a result, the design data AI sees is most likely graphic-design data, not web-design data.
Bai Ya says, now you can tell whether a web page is Vibe Coded just by checking whether its default title is Songti.
This was a scene on Wu Xiaobo Channel's "AI Future Talk" livestream. The host was our Unique Research founder Wu Wei. This session's topic was "in the AI era, has the one-person company truly arrived?" Over one conversation, Bai Ya tossed out many counterintuitive judgments.
The term Vibe Coding was coined by OpenAI co-founder Andrej Karpathy and was Collins Dictionary's Word of the Year last year. In YC's latest winter batch, a quarter of startups generate 95% of their code with AI. Tools like Cursor and Lovable already let people who can't code "speak a product into existence." Lovable's ARR has already hit 100 million US dollars — the market is voting with money.
The question itself is harmless; it's one line of CSS. But it punctures something: when handling the emotional, aesthetic, and tacit industry knowledge, AI is like an honor student who read many books but never actually "did it by hand." It knows all the rules, but not the "craft" that was never written down.
Behind this small matter hides a bigger question — in the AI era, how much are the experience and expertise we're proud of actually worth?
The one-person company isn't the endpoint. Eliminating yourself first is the starting point.
Your Experience May Be Holding You Back
Bai Ya says his biggest lesson these two years has been overcoming inertia.
What was his past superpower? As long as you described the user scenario and need clearly, an image quickly surfaced in his head — how to lay out the core interface, how the flow goes, how to design the interaction, and it was roughly optimal. This intuition is muscle memory built over twenty years of design plus product.
But this muscle memory became a handicap in the AI era.
Why? Because he was too used to "getting it right in one step." The first image that popped into his head was usually the interface, a concrete interaction form, not the underlying logic. When describing needs to AI, he'd unconsciously shove this image straight at the model — "you should lay it out this way, interact that way" — and what AI produced was merely a replica of his old experience, with neither transcendence nor surprise.
Vibe Coding doesn't put programmers out of work. What it truly changes is the starting point of product design: from "draw the interface" to "say the need." Bai Ya used a metaphor: in the past you wrote the exam paper for AI to solve; now you tell AI what you want tested and let it set the questions, solve them, and grade itself.
This shift sounds easy but feels awkward. Like a veteran who drove manual for twenty years suddenly put in the passenger seat, moving only his mouth and not his hands, feeling off all over. Bai Ya admits these two years he's mainly been doing this — restraining the urge to "jump straight to the interface," pulling thinking back upstream: what problem does the user actually want to solve, what's the mindset behind it, where's the core constraint of the business logic, then feeding these to AI to generate concrete options.
He also found an interesting phenomenon: people who used to be bad at design may actually build better products with AI — because they don't hold so many "I think it should be this way" obsessions, and instead let AI experiment, iterate, and surprise.
He added a key line: "you must believe AI is smarter than me."
Don't rush to dismiss this. Think it through: if you hard-code a Prompt — "never use Songti" — AI won't repeat the mistake, but it also loses an ability: using Songti well in scenarios where Songti is right. For example on some clothing-brand sites, Songti just fits better than heihei. Hard-coding rules is personally castrating the model's generalization ability, and generalization is precisely the source of AI creativity.
Preserve AI's generalization ability, and you preserve its creativity. Your experience shouldn't be an instruction; it should be context, background, material that helps it understand "why," not a checklist of "how."
Veterans with ten-plus years feel this by now — the thing you're best at may be becoming your biggest resistance in collaborating with AI.
Will Companies Disappear? No, But Roles Will Merge
The industrialization arc is familiar: assembly-line mass production → disciplinary education → everyone trained into a screw → roles get finer and finer → collaboration costs rise exponentially → management science is born → overall efficiency drops instead.
The internet industry replicated this logic. A product one person could once write was later forcibly split into market research, user research, product features, interaction design, UI, front-end, back-end, database, ops... twenty roles. Each role got more specialized, but the cost of communication, alignment, and bickering rose with it.
Bai Ya's judgment: the road of ever-finer social division of labor has reached its end; what's next is merging.
Not everyone has to run a one-person company — that's the extreme form. More realistically, enterprises will see many "one-person roles." One person, simultaneously product manager, UI designer, front-end engineer, using AI tools to get all three done together.
Youzan itself is a pioneer. They manage a sales network of nearly two thousand people; this sales-management system used to quote a million RMB a year for external customization, with an in-house team of 10–15 people costing 5–6 million a year. Later Bai Ya assigned two people — one product, one engineering — using superun (the conversational-programming product he founded for non-technical people, now his main bet), and in under two months cobbled together a complete system. Presale smart suggestions, in-visit recording review, sales-lead routing — everything you'd expect was there.
Total cost dropped to 2% of before, and the management-software procurement budget went straight to zero.
Two people, two months, replaced a team. This is no longer the rhetoric of "cost reduction and efficiency"; it's real arithmetic.
This case reveals a bigger judgment. Bai Ya splits enterprise software cleanly: one category is outward-facing marketing and sales systems that must innovate in real time with market changes — buy platform services for this; let professionals chase the trend and the feature race; don't mess with it yourself. The other is inward-facing management systems, stable in process but extremely idiosyncratic — every company's expense approval, customer visit, lead-routing logic differs; these are best cobbled yourself with Vibe Coding, because only you know the real internal process and pain points; buying outside SaaS always means cutting your feet to fit the shoes, while self-built is more fitting, more flexible, and cheaper.
At bottom, you didn't self-build before because the cost was too high — building a team, hiring developers, managing projects, burning millions a year. Now two people in two months ship output; marginal cost suddenly hits zero, and the math changes entirely.
The one-person company isn't everyone becoming a lone hero. But in every company, there will be roles where "one person is a whole department." This change may come faster than most people imagine.
The Real Bottleneck of AI Transformation Isn't Technology, but the Boss Not Using It Themselves
After role merging and organizational change, many readers may ask: so how should my company start AI transformation?
Bai Ya gave a counterintuitive answer: if the number-one person doesn't deeply use AI, transformation can't take its first step.
This sounds harsh, but on reflection it's true. By business intuition, what's the standard move when an enterprise embraces new tech? Get traffic, drive acquisition, do growth — these are the scripts bosses know. So you often see this scene: the boss sees on the news that AI is changing the world, slaps his thigh and says "we must embrace AI transformation," then turns and tells subordinate Xiao Wu to do it. Xiao Wu gets the task dazed, worrying: "if I pull this off, am I the one getting optimized?" Reluctantly passes it to Xiao Li, who half-heartedly tries Doubao and reports back "AI is nothing special."
A mighty AI transformation dies on the third layer.
Bai Ya gave a precise analogy: a boss needn't use Photoshop, but must understand aesthetics; needn't write code or tune agents, but must walk the path themselves and know AI's capability boundary and feel. A company whose number-one person isn't a deep AI user can't take even the first step — because you don't know which direction to send the team.
But even if the boss uses AI, a bigger trap awaits: internal enterprise knowledge governance is a mess beyond imagination.
Bai Ya told a real case. A hardware product released version 1.0; the R&D doc wrote plainly: "black-gold luxury black edition," under 300 grams, 10-hour insulation. The Tmall team took it and unilaterally renamed it "hot-gold deluxe edition," changed "cold retention" to "over-10-hour warmth," made "under 300g" precise to "289g" — that became 2.0. The Douyin team edited again, producing 3.0. Finally three versions of docs lay internally: PR uses 1.0, Tmall uses 2.0, Douyin uses 3.0 — total chaos.
Even people are dizzy; feed this messy data to a large model, how can it not be confused?
So Bai Ya says, in the past we talked "data operations"; now we must move toward "knowledge governance" and "knowledge operations." Youzan has already created a "knowledge operations specialist" role, whose core duty is having the large model auto-mine knowledge points and auto-search-and-alert one-channel changes across all channels. Before your internal enterprise knowledge is sorted, don't rush to AI — Garbage In, Garbage Out; this truth gets amplified a hundredfold in the AI era.
Take the most intuitive AI-landing scenario: customer service.
In the past everyone thought AI customer service just saved a few human costs. Outsourced service reps earn 5–6k a month, 85% accuracy in week one, at most 95% after a month of training. AI service accuracy hits 98% directly, at very low cost — but that's only the surface benefit.
Bai Ya says after Youzan launched AI customer service, it found three bigger incremental values. First, refund and complaint rates fell markedly, because AI is always in a great mood, online 24 hours, and never talks back when scolded. Second, customer service turned from a cost center into a "sales center" — AI can digest all new-product data in a second, precisely cross-sell, push B and C when someone buys A, and incubate sales leads. Third, the role walls between service, sales, warehouse, and R&D broke down; information began flowing freely inside the org.
Getting to this point, Bai Ya tossed out an even bolder judgment: AI can't replace the CEO, but it can replace more than half of the CEO's repetitive work.
He himself is an ENTJ, a classic "super producer." What used to frustrate him most? Spending most of his energy on tedious communication, coordination, and meetings with hundreds or thousands of people; "organizational consumption" far exceeded doing the work itself. Now with AI as the super productivity tool that decouples human labor, he can finally put time back on the work itself — so he says he's "not anxious, incredibly excited."
What Should Ordinary People Do? Two Concrete Suggestions
If you've read this far thinking "I get the logic, but what do I, an ordinary person, do?" — this part is for you.
Bai Ya gave two very concrete suggestions, not chicken soup, but actions you can start today.
First: build the muscle memory of "ask AI anything, anytime."
On any question in life or work, pull out the phone and speak Mandarin into it directly; don't type. Treat AI as part of your body and the entry point to everything. Think of your old search-engine flow: distill into keywords in your head → a list full of ads comes up → you sift, read, and summarize yourself. Now? Natural-language dialogue, a photo, directly a summarized native answer. If you can't even build this advanced lifestyle, you can't use AI well at work.
Second: find an AI solution to eliminate yourself at work.
If you're a product manager, copywriter, or analyst, don't just want AI as an assistant doing odd jobs. Truly smart and extreme people should proactively train and build an AI clone or agent workflow, replacing repetitive, procedural, non-creative work 100%. When you can use AI to eliminate your current self, you've hit a real career-level jump.
After suggestions, let's talk about anxiety. Because this era lacks no suggestions; what's lacking is the mental energy not to be drowned by anxiety.
Bai Ya has seen too many peers this year: up all night deploying and tuning open-source models, chasing what model updates today and what tool comes out tomorrow, afraid to lag half a step. Many super individuals, once on Vibe Coding, want to do more and more, ending up busier and more tired than before.
His own strategy this year runs the opposite way: restrain himself; let a new AI product make noise for three or four weeks before he deeply tries it. Let others pay the cost of trial and error and scout the path first, then he follows.
He used a metaphor I find especially apt. The old anxious feeling was like waiting on a platform for a high-speed train, not even daring to use the bathroom, afraid that while you're gone the train arrives and you miss the whole era. But after realizing one thing, he relaxed — the absolute theme of the next ten years or more is AI, so go use the bathroom in peace. Catching it a day, a month, or half a year later makes no essential difference against an era train long enough.
More importantly, human-brain growth and the increase of compute and mental energy must be nourished by enough sleep and real life. Staying up late tuning models and fiddling with tools every night, your bottom-layer compute the next day is definitely deficient. Without real life, mental energy collapses and creativity dries up. Conversely, with AI tools taking over meaningless repetitive labor, take time back to real life — cycling, walking, shows, meditation, hanging with friends — refill your abundant mental energy in real life, and only then, going back to products and CEO-ing, does creativity truly become powerful.
Coda
Near the end of the interview, host Wu Wei asked Bai Ya a question: once a person has AI, is he still simply a "human"? Is he a composite of "human + Agent"?
Bai Ya thought and said: call it a "superhuman." Humans were always ahead of other species for one reason only — using tools. The agricultural era used hoes and shovels; the industrial era used machines; today machines think, so we use AI. Fundamentally it's still constantly using more advanced tools. To the user, AI is a tool that augments himself.
So, don't be anxious. AI hasn't turned humans into something else; it just upgraded your toolbox. You'll still be hungry, sleepy, needing the bathroom — these human bottom-layer operating systems haven't changed at all.
How to live next? Bai Ya already gave the answer: eliminate yourself first, then go use the bathroom when you need to.
More Conversation Detail
Core-view section: the essence of AI and the businessman's view
Q1: Wu Wei: Mr. Bai, lately everyone talks about super individuals and one-person companies (OPC). In the AI era, if one person uses an Agent to do design, customer service, and sales all, is he still a "human" in the pure sense?
Bai Ya: Call it "superhuman" or "super individual," either works. But don't overcomplicate it; humans were always ahead of other species, core because we use tools. The agricultural era we used hoes and shovels; the industrial era we used machines. Today machines think, so we start using AI. Fundamentally, AI is still a more advanced tool that greatly augments your own capability.
Q2: Wu Wei: Many tech companies are desperately racing on large models and underlying tech; Youzan is so big — haven't you thought of starting a large-model company?
Bai Ya: One critical thing in being a person is "knowing yourself"; never lie to yourself. Now 99.9% of the world's resources and smart people are grinding on large models; that needs extremely foundational, passionate, tech-fluent scientists doing basic research. I'm trained in fine art; I don't understand underlying tech, so doing that would be pure self-deception. What I can do, love, and am confident doing well is the application layer — orchestrating all the models together, building good products end-users can directly use based on real user scenarios.
Q3: Wu Wei: You said at the opening that your "excellent product ability," the thing you were proud of in the PC and mobile internet eras, has instead become a "handicap" in the AI era? How to understand?
Bai Ya: Over the past two or three decades doing product I had a strong inertia: once you described the user scenario, an image immediately popped up in my head, extremely fast in drawing the core interaction interface, UI, and usage flow.
But in the AI era this is a huge handicap. Because I jump too fast, going straight from logic to the concrete interface. The core of using AI well is absolutely not dictating to the model "what you must look like, how you must work"; it's fully feeding it your core needs, thinking path, and context. It took me nearly two years to throw away this overly concrete "super ability" and learn to converse with the model with a "new head."
Aesthetics and product section: the "most expensive ability" of the AI era
Q4: Wu Wei: Lately large-model auto web pages (Vibe Coding) are hot, but you complained on your Moments that any Chinese web page and app Vibe Coded has Songti as the default title — is this an industry secret?
Bai Ya: This is classic "the large model didn't see all of humanity's hidden crafts." Those of us from web-design backgrounds, the first class we learned surely had an iron rule: Chinese web pages absolutely can't use Songti. In the early days browsers rendered poorly; Songti's elegant sharp corners were all jagged on screen, very uncomfortable, so Chinese web pages default to rounded, regular heihei.
But AI scientists basically don't care about this aesthetic. When training, AI saw countless code and graphic-design books. Graphic design (like books and posters) often uses Songti on big titles to lower the image's aggression. AI mistook this for the universal standard of Chinese design, so when it Vibe Codes, all the web titles come out Songti. Now, to tell whether a web page was directly cobbled by AI, check whether its title is Songti — spot on.
Q5: Wu Wei: Since you found this AI quirk of liking Songti titles, why not just hard-code a prompt in the backend, "never use Songti," and be done?
Bai Ya: That's the worst, poorest way to use AI — you always think you're smarter than AI, so you contrive to hard-code rules and limit it. Once you hard-code it, you not only stifle AI's powerful generalization and intelligence; more frightening is, when the next large model upgrades, you enjoy none of the systematic upgrade dividend.
Moreover, there really are scenarios where Songti is better — for example a high-end clothing brand's homepage opens to a full-screen big image paired with elegant Songti, very on-tone. So superun's rule is: absolutely no hard-coded limits; instead spend effort feeding it aesthetic philosophy, context, and awareness to guide it. Let it output heihei when heihei is called for, and only in the rare truly fitting, stunning scenario let Songti emerge as an Easter egg.
Q6: Wu Wei: In the AI era everyone has equalized productivity tools; an ordinary person can crank out a 60- or 70-point design. So how important is "aesthetics," this elusive thing, in the future?
Bai Ya: Aesthetics is humanity's "most expensive ability" in the future, bar none. When social per-capita GDP exceeds 30,000 USD and routine labor is all taken over by AI, material abundance makes human demand for spiritual civilization, art, and sports expand infinitely.
If everyone finds mediocrity acceptable and everything made looks identical, then the aesthetic people stand head and shoulders above. Because what we ultimately produce is for human eyes; humans are naturally willing to linger and glance again at beautiful things.
Q7: Wu Wei: So as an ordinary white-collar or enterprise boss, how do you quickly raise your aesthetic ability in the AI era?
Bai Ya: My advice may be counterintuitive: don't try to raise your aesthetic ceiling; that'll be very frustrating. What you really must do is firmly hold your aesthetic "floor."
I've seen many middle-aged straight-male bosses in traditional industries whose product packaging, logos, and storefronts are truly ugly, but he's utterly unaware. He keeps wanting to spend big to hire a design director with ultra-high taste to gatekeep; I tell him directly: if your own aesthetic floor is low, no matter how great a master you hire, it's useless. From now on, any design coming out of the company that you find merely "passable, OK" — kill it outright; only when you see it and think "wow, truly stunning" do you let it pass. Take your ceiling as the company design's lowest floor, and aesthetics naturally rise.
Organization and software-industry section: from incremental consumption to a 2% cost disruption
Q8: Wu Wei: You proposed a "role merging" concept, thinking future social division won't get finer but coarser, the opposite of management over the past few hundred years?
Bai Ya: Past centuries of industrial civilization all pursued fine assembly-line division of labor, training people into screws in one role. This leads to more and more roles; with more roles, internal communication and collaboration costs rise as exponential entropy, and overall efficiency actually drops. To build software used to require market research, user research, product features, interaction design, UI, front-end, back-end, ops... work one person could do became twenty people meeting to align.
After AI appeared, it directly integrated this middle layer of mental division of labor. The workflows and projects may still be those 20, but roles can merge at scale. Product manager, UI, front-end in one — one person closes the loop end to end.
Q9: Wu Wei: Sounds disruptive. Does Youzan internally have a real app built through such "role merging" running on core business?
Bai Ya: Yes. The whole extremely complex "sales-management system" managing our nearly 2,000-person sales network was directly cobbled by one product manager plus one engineer over a month-plus on the superun platform.
Q10: Wu Wei: A system cobbled by two people — can it compare to mature CRM software on the market costing hundreds of thousands or millions? How fine can it manage?
Bai Ya: It's far more fitting and usable than generic off-the-shelf software. It's fully customized to our company's idiosyncratic management philosophy.
For example, when our salesperson visits a client, they pull out the phone and tell the system what type of client they're visiting; AI automatically pulls cases from the library and in one second pushes core selling points, scripts, and pitfall guides to the salesperson.
During the visit, full recording; that night in my Hangzhou office, AI has already listened and clustered all 800 frontline visits nationwide that day and produced a complete daily summary: where clients' objections that stopped them from buying cluster, which selling points land best, on which line our salesperson couldn't answer. Whatever question I'm curious about, I click in and hear the real on-site audio clip. This directly guides next-day product refinement and employee training.
If a salesperson just signed a chain pub, one sentence in the system and AI immediately washes out the 20 most precise nearby pub leads from the public lead pool and assigns them, letting him strike while hot.
Q11: Wu Wei: This system built by two people — versus past in-house or bought standard software, how did cost change?
Bai Ya: Before, to do this we either paid an external software company million-level service and customization fees a year, or kept a high-salary in-house team of 10–15 people, burning 5–6 million a year just on salaries, uneconomical.
Now we finished development with two people; after it's fully stable by this year's Q3, the product and engineering people directly move off to other work. Going forward this system only needs one-third of an ops person plus half an engineer for daily maintenance, running in the cloud with no high concurrency; compute and Token cloud service costs are only a few thousand a year. Total cost dropped to about 2% of before! So now our internal management-software procurement budget is written as "0"; no more procurement at all.
Q12: Wu Wei: You took the procurement budget straight to zero — how will those traditional software companies and SaaS companies survive?
Bai Ya: Software must be split into two categories.
The first is outward-facing incremental systems directly tied to big marketing/sales/channels/traffic. These systems' platform rules (WeChat, Douyin, etc.) change every day, and market playbooks iterate in real time. If your enterprise locks the door and cobbles with AI, your iteration speed can't keep up. These systems will fully move to shared enterprise-service platforms; you must use their service ecosystem to keep pace.
The second is inward-facing internal management systems with extremely stable processes but highly idiosyncratic needs (like expense, evaluation, HR). People bought standard SaaS before because customizing was too expensive and they had to compromise. Now self-cobbling management systems with AI costs one-tenth or even one-hundredth of before. More and more enterprises will choose AI-customized self-built. Traditional standardized software suites will gradually give way to more flexible self-built options, and software companies' roles will evolve from selling a fixed product to providing deeper industry practice or consulting.
Landing and pitfall-avoidance section: how to take the first step of AI transformation
Q13: Wu Wei: Many traditional enterprises or SMB merchants have very limited budget and energy. With AI arriving, should they first use AI to solve acquisition/traffic problems, or first do customer service and save money?
Bai Ya: By normal business intuition, everyone wants to start with traffic and acquisition. But in fact this often fails; opening up the top of the funnel counterintuitively fails at the first step.
Chinese enterprises have a severe "paternalistic" management tradition. Often the boss sees on the news that AI is changing the world, slaps his thigh: "we must do AI transformation! Xiao Wei, you handle this." After giving the order, the boss never touches it. The white-collar worker Xiao Wu doing the work is also dazed, even thinking if this works, might he be unemployed? The white-collar passes it to the intern, who finally half-heartedly tries Doubao and reports AI is nothing special, and the boss hears that and drops it. A company whose number-one person isn't a deep AI user absolutely can't take the first step of AI transformation. A boss needn't write code, but must walk the path himself and have the "feel" and "body sense" of AI's capability boundary.
Q14: Wu Wei: After the boss has the feel, when an enterprise truly introduces AI landing, which part do you think shows results fastest and should be onboarded first?
Bai Ya: When AI employees onboard, what shows results first is always AI customer service and AI sales. We used to think like many merchants, that introducing AI customer service was just to save two outsourced reps' wages and raise accuracy a bit.
But later we found AI brings far more than cheapness:
First, it has extremely low refund and complaint rates. Outsourced reps under angry, massive repetitive inquiries have emotions and slips. AI is always 24/7 full marks in attitude, never talks back when scolded, no negative emotion; the store's return/exchange and complaint rates slide straight down.
Second, it turns customer service — traditionally a "cost center" — into a directly revenue-generating "sales center." Before, asking a few-thousand-yuan outsourced rep to memorize all the selling points, specs, and pairings of 50 new products this week and actively cross-sell was impossible. But AI can digest full inventory and new-product data in a second; when a user buys A, the chatbox immediately recommends B and C that can go along, and very keenly incubates high-AOV sales leads in the conversation. Lower cost, better attitude, and it grows sales along the way — that's the real first step.
Q15: Wu Wei: Sounds perfect, but isn't there a major hidden premise: enterprises must organize their internal data, SOPs, and docs extremely standardly to feed AI, otherwise AI talks nonsense?
Bai Ya: Exactly; this touches every enterprise's biggest bottleneck now: from the old "data operations" to fully moving toward "knowledge governance" and "knowledge operations."
Now many companies' internal knowledge bases are a total mess. R&D launches a new product; the 1.0 doc calls it "black-gold glossy-black edition," under 300 grams, 10 hours insulation/cold retention. Within two days of launch, Tmall ops, grabbing attention, privately renames it in the backend to "hot-gold deluxe edition," finds "cold retention" too obscure and changes to "over-10-hour warmth," finds the weight too vague and writes weighed "289g" — producing a 2.0 doc. Two days later Douyin livestream changes a bunch more words in the frontend for gameplay, producing a 3.0 version.
Three dead docs lie internally; even insiders are thoroughly dizzy. You now feed this conflicting, fighting data straight to the large model — how can it not be confused? In the past, hiring someone to keep R&D, e-commerce, livestream, and PR all real-time aligned was nearly impossible. Now through AI's "knowledge operations" tools, AI can in the backend auto-discover, mine, and proofread this chaos and alert sync. Turning information that used to be dead on paper and dead in different departments' docs into a living, highly consistent asset — that's the infrastructure.
Individual philosophy and life section: don't burn out your bottom-layer compute and mental energy
Q16: Wu Wei: For ordinary white-collar workers or individuals without a tech background, wanting to stay "advanced" in the AI era, what concrete action guide do you have?
Bai Ya: Two most practical suggestions.
First, in life and work, thoroughly build the muscle memory of "ask AI anything." Whatever life chore or workplace problem, pull out the phone and speak Mandarin directly into WeChat AI, Doubao, or Qwen; never type. Before, with search engines you had to filter and compute the question in your head into a few "keywords," type them, and sift and summarize the ad-laden list yourself; the bar was too high. Now just jabber in your most natural mother tongue, and it feeds you the distilled native answer directly. If you can't even build this lifestyle, your consciousness is already behind.
Second, truly diligent and excellence-seeking people should now find an AI solution to "thoroughly eliminate themselves" at work. If your current job daily does only routine, repetitive, fixed-logic things, don't just think of AI as a little copywriting assistant. You should proactively train an Agent as your clone, letting it 100% take over your current repetitive work. When you can successfully use AI to "eliminate" your current self, you've completed a real career-level upgrade.
Q17: Wu Wei: Many super individuals, after using Cursor or OpenClaw ("little crayfish"), find they want to do more and more, ending up busier, more tired, staying up all night, extremely anxious. How to break it?
Bai Ya: My action guide for myself this year runs the opposite: I only touch a new AI product after it's been hot and popular on the market for three or four weeks.
Many peers' anxiety now is like concentrating on a platform waiting for a high-speed express train, not even daring to use the bathroom, afraid that while you're gone the train comes and goes and you miss the era entirely. But if you realize the absolute theme of the next 10 years or longer lifecycle is AI, then go use the bathroom in peace, rest assured! Catching it a day, a month, even half a year late, against this long enough era train, makes no essential difference; everyone ends up on the same train; you miss nothing. Stop staying up all night taming little crayfish.
Q18: Wu Wei: Sounds counterintuitive. As a "super producer," you're actually telling people to fiddle less and follow trends less?
Bai Ya: Human-brain growth and the increase of compute must be nourished by enough sleep and real life. If every night you don't sleep, fiddling with tools and extremely anxious, your next day's bottom-layer compute is definitely severely deficient.
If a person has only work and no life all day, your "mental energy" also collapses, and in the end your creativity dries up completely. With handy AI tools taking over non-creative repetitive labor, we should take time back to real life — cycling, walking, shows, meditation. Refill your "compute" and "mental energy" in real life, and only then, going back to products and decisions, does your creativity truly explode.
Q19: Wu Wei: Youzan is deeply integrated with the WeChat ecosystem; WeChat recently launched native WeChat AI interaction. Do you think this is a super huge new trend for content creators or ordinary individuals?
Bai Ya: In my heart, WeChat has two extremely deep, bone-deep tags: first "decentralization," second defining itself as "a tool striving for all users." Moments, Official Accounts, Mini Programs all inherited these two core genes. And I think Channels is somewhat in friction with WeChat's genes, because it's too centralized.
But this time WeChat's native AI perfectly fits both "tool" and "decentralized." This AI runs inside WeChat; it knows each of us; it's the super tool that truly handles our personal business. This is like taking the decentralized power of Official Accounts plus Mini Programs and wildly amplifying it again through AI. The room for small businesses or individuals to do lightweight copywriting, services, and high-efficiency connection through WeChat AI is enormous.
Q20: Wu Wei: At the end, Mr. Bai, please use one line to set an example for all the ordinary people in the livestream, still lost and anxious, looking for AI opportunity.
Bai Ya: If you've taken a personality test and found you're at heart a "producer" who wants to make and tinker with things, then in the age when humanity's most powerful super-production tool (AI) ever arrives, you need no anxiety at all; you should be utterly excited! Hand the heavy physical labor and non-creative chores fully to tools, restrain the urge to open more projects, sleep well, live real life well, then pour all your compute and mental energy into that one creative thing you truly love!