Original · Unique Research · 2026-06-08
Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the four-panelist AI-content roundtable (Qiyu Shuke / Shen Yang, Ji Su Engine / Wang Jun, Guomai Zhiyu / Li Deheng, Yunxigu / Luo Hui, hosted by Wu Shenliang), the five themed sections, and the full transcript. All named companies, products, people, and figures are preserved. Founder and panelist statements are source attributions, not independently verified findings.
AI Industry Observation
Who is tightening the screws on AI content's assembly line?
AI content's factory has started up, but the assembly line isn't built yet.
"
We've only just reached the moment of having electricity and machine tools.
A 20-person company, annual revenue in the tens of millions, with a hit title whose plays across platforms broke 100 million.
Another serves over 60% of China's auto dealer groups; clients need to do nothing at all — AI shoots their videos, runs their livestreams, places their ads, replies to their messages, and delivers sales leads straight to them.
A third ships 10 million units a year of children's smart hardware and is now using foundation models to batch-produce localized story content for kids around the world.
It sounds like three unicorns. In reality they add up to maybe fewer than 200 people.
But sitting at the same roundtable, these founders drifted toward a surprising consensus: AI content's factory has started up, but the assembly line isn't built yet.
"We've only just reached the moment of having electricity and machine tools," said Luo Hui (罗辉), founder of Yunxigu Technology (云希谷科技).
That line set the tone for the whole discussion.
The factory has started up
First the good news. AI content production is no longer slides and proof-of-concept; several companies are already making real money from it.
What Shen Yang (沈扬) of Qiyu Shuke (企域数科) does is probably the most "hardcore" of the room — using AI Agents to run the full new-media marketing chain for auto 4S dealers. His description is a little startling: "The client doesn't need to do anything; they just cooperate with us on some initial device setup. After initialization, for everything like going live and short-video production, including ad placement, the client needs no involvement; we directly deliver sales leads to them."
Translate that: after a 4S store connects to the system, AI auto-generates short-video scripts, auto-edits, auto-schedules livestreams, auto-replies to barrage and DMs, auto-runs ad placement, and what lands on the store's desk is a list of customer phone numbers to follow up on. As for people? People go do something else.
Qiyu Shuke took Sequoia China Seed Fund's angel round and a Pre-A led by SIG with Tencent following, and has served over 100 auto-brand OEMs. Shen Yang previously managed digitalization for 140-plus 4S stores, so he knows these stores' pain points: "They have no mature hosts or video shooters/editors; it's hard to hire and hard to keep them."
Ji Su Engine's (极速引擎) founder Wang Jun (王俊) took a different path; he does AI short dramas and comic dramas. This Nanjing company has only about 20 people but is already a core head content partner across major platforms, with a hit title breaking 100 million online. It was selected into Google's Going Global Accelerator — the only AI content-operations company in it.
Wang Jun put it directly: "Our company is now basically semi-automated; everyone is all-AI, and there's almost no step where AI isn't present."
The trump card of Guomai Zhiyu's (果麦智娱) co-founder Li Deheng (李德恒, Henry) isn't technology but the group's deep IP resources. Parent company Guomai Culture (果麦文化) is an A-share-listed enterprise, deep in cultural media for over a decade, holding exclusive signed copyrights of head authors like Yi Zhongtian, Luo Xiang, Han Han, and Cai Chongda, and co-producing theatrical films like The Continent and Pegasus. Guomai Zhiyu is precisely the group's key vehicle for monetizing installed and original IP and laying out AI-visualized content development and commercial operations.
Yunxigu's Luo Hui is the only one doing hardware. This national-level "little giant" specialist-innovative enterprise ships 10 million children's education smart devices a year. Luo Hui calls hardware "bowls and chopsticks" and AI education content "the meat and greens" — "we just provide a better tool to make it easier to absorb this educational content."
Four companies, four paths, but one common trait: they're already making money with AI. This isn't a discussion of feasibility; it's a discussion of yield rate at scale.
The machine tools have a fault
Good news done. The bad news: full automation isn't happening any time soon.
Wang Jun's answer on this was the most technically insightful in the room. He didn't follow the "AI keeps getting stronger" narrative; he directly questioned the underlying training paradigm:
"AI's underlying training is: tell AI humans like this thing, and it produces this thing. But 'what's right and what we need' and 'what humans like' are different; what it produces is probably more of an average."
This is worth a second thought. A model trained by RLHF (reinforcement learning from human feedback) essentially approximates the statistical center of human preference. But good content is precisely not "an average" — it's the outlier that makes you stop in the feed. AI naturally leans toward generating things "everyone thinks is fine," not things "that make a certain group addicted."
Shen Yang added another dimension from the front lines: "How to avoid becoming homogeneous content, while constantly keeping up with platform algorithm trends to produce hits or premium content — that takes work, it needs human judgment and coordination."
A trade secret of the industry is hidden here.
Between platforms and AI content producers is essentially a cat-and-mouse game. Douyin's rules are clear: AI-generated content must be labeled, and low-quality homogeneous content gets directly throttled. TikTok and YouTube are the same. 2026 Q1 data shows domestic AI short dramas already reached 95%, so homogeneous that platforms had to step in.
But throttling isn't aimed at the "AI" label itself; it's aimed at content quality. The "black-box evolution" Shen Yang mentions is actually a survival strategy — since the platform algorithm is a black box, use AI to run low-cost massive A/B tests and grope for the parameter tipping point that avoids throttling and still gets traffic. "The parameters that work for our content in the Sichuan-Chongqing region, the Northeast, and East China are completely different; we fully follow the A/B test results."
In other words: if the platform suppresses v1.0 pure-AI garbage, content companies must use AI to evolve a v2.0 within 24 hours. This isn't feeding garbage; it's high-intensity adversarial training against the platform algorithm. Whoever grabs 5% more traffic than peers in this dynamic game survives.
Luo Hui positioned this dilemma with his industrial analogy: "For brainwork products, consistency out of the 'machine tools' is currently worse; people still have to do re-inspection. Whoever builds this assembly line most perfectly will lead the whole market."
The most expensive part on this assembly line, for now, is still the human brain. AI is the electricity, the machine tools; but the quality inspector still has to be a person.
After production cost hits zero, what's most valuable?
The host threw out a sharp question: AI has exploded content supply, but user attention hasn't increased. When production cost drops to near zero, what is the industry's scarcest thing?
Four people gave four different answers. Interestingly, the four answers together form a complete competitiveness puzzle.
Shen Yang says it's aesthetics. "All good content, marketing or entertainment, gives people aesthetic and spiritual enjoyment. In the end it's about the team's cognition of beauty, and your ability to industrialize the production of beauty." He gave the example of the original iPhone — "you define it, turn it into a standardized product; that's actually a very strong ability to understand beauty, and to break it down into something engineering-able."
Wang Jun says it's the team. But the team he means isn't "a few talented people," it's an un-exportable organizational form: "We aren't afraid of people being poached, because poaching a director from me to your company may not fit. Our org structure, role naming, are all different from traditional film companies. The production method is completely different."
Li Deheng says it's original worldview and emotional resonance. "How to keep producing quality products with a more original worldview — those true content-creation sources that have thought and value. Second, the ability to precisely grasp emotional resonance across different demographics."
Luo Hui translated this into eating: "Today in Shenzhen everyone eats quite full. How do you make consumers like eating yours? First, your ingredients must be relatively scarce. Second, you can provide more personalized service."
Investors will surely press: do these "moats" hold up when big companies come crashing in?
I think the answer is hidden in a classic paradox of business history — the innovator's dilemma. Why don't big companies come grab it? Not that they can't, but the math doesn't work. Shen Yang does new media for 4S stores; Wang Jun does niche-IP comic dramas. These vertical tracks, in the eyes of a big company with tens of billions in annual revenue, are "too little fat, too hard bone" chickens ribs. A big company's billion in funding would rather burn on general foundation models and compute clusters than send a highly paid team to teach auto dealers how to mount cameras every day.
A vertical company's real moat is the profit margin big companies disdain, and the agile grind big companies can't learn.
Building an overseas branch: who decides?
When the topic turned to globalization, Wang Jun's tone changed noticeably: "Before, shooting a live-action drama, 1.5 million RMB got you into the domestic game table, and going overseas for distribution cost another $200,000 to $300,000 to reshoot. Now it's totally fine; go ahead and make it, all the minor languages we didn't dare touch before are doable."
AI has indeed smashed cross-language content production cost to the floor. But between "can make it" and "can sell it" stand a wall of culture and a wall of regulation.
On the culture wall, the four had subtle disagreements.
Wang Jun's formula was cleanest: "Production must be global, but the consumption side is definitely still local; this is unavoidable."
Luo Hui was more optimistic, giving a specific ratio: "80% may be globally universal, but 20% may be special local content." He's already doing it — Yunxigu's children's story content "ships in volume to every country globally," regenerated by foundation models based on each country's culture.
Li Deheng was cautious. He agreed technology and channels can achieve globalization, but stories and content carriers are rooted in local literary lineage and hard to strip of cultural identity. On that logic, Guomai's going-global takes a steady approach, first picking historical and pop-culture themes, using Hong Kong as a hub to test the waters.
Shen Yang applied his "black-box evolution" logic straight to going global: "In the AI era we can do black boxes. Today's algorithm platforms give clear data feedback on exposure-to-conversion for every piece of content; we can use AI to build workflows, run low-cost massive-testing black-box evolution, and output content that truly fits a given country or region."
Read these judgments together and they converge on one sentence: there's no longer a technical barrier to globalization on the production side, but localization on the consumption side still needs people (or at least AI people have trained) to hold the gate.
But there's an elephant in the room no one mentioned — regulation.
2026 is the key year the EU AI Act takes broad effect. Transparency obligations have started; high-risk-system compliance deadlines are pushed to end-2027, but the fine cap is set — 7% of global turnover. US state AI legislation is accelerating too. Multiple Southeast Asian countries have strict local review of foreign digital content.
The domestic playbook of "AI mass spinning, matrix bombing" won't work in overseas markets. Whoever treats AI as a cheat to dodge regulation dies on the beach.
The truly viable path is turning AI into a compliance filter — using AI to auto-detect imagery that offends local cultural taboos, auto-stamp traceable watermarks on content, auto-adapt labeling requirements for different countries. Li Deheng's and Luo Hui's going-global practice is essentially this: using China's mature IP and hardware supply chain as the base, letting AI play translator and localization lubricant.
What will this assembly line look like in a year?
The last question was predictive. Four people, from different angles, drew the same picture.
Luo Hui started from the cost end: "Writing software to make content used to cost a million; in the future it might cost 100 RMB." His inference: product sales will basically become free, shifting to post-pay by usage. Content's business model gets rewritten.
Wang Jun judged from the form end — and his timeline was more aggressive than everyone's: "Maybe in half a year it changes a lot." He raised an interesting hypothesis: "Could there be a tool or app where, the moment you finish content, it's already distributed and being watched? Maybe there's no such thing as a 'distribution' step anymore." Production and distribution merge into one; the middleman's space is fully compressed.
He raised another, perhaps deeper direction: per-person content. "We 100-plus people here watching the same drama might not be watching the same thing, but underneath it's one IP. I might like the cute type, he likes the elegant type, what you watch is totally different from mine, yet we both love it."
Li Deheng's angle was unique: he believes AI digital-cultural content will gradually evolve into a tradable standard asset. "As AI in content creation keeps spawning new roles, new plots, worldview settings — these are highly valuable new content assets. In the future this could spawn platform-type enterprises deep in IP rights confirmation, asset flow, and underlying infrastructure."
Shen Yang's prediction was most down-to-earth and vivid. He said internally they call coding without AI "old-school coding." Then he described a marketing future: "Maybe every salesperson or store will have an AI-built digital IP, becoming the perfect person in your mind. Able to sing and dance, able to present the product." He called it "professional beauty filters."
"Most people can't stand the workload pressure of posting daily and livestreaming, but AI can turn a living person into the most perfect digital IP."
Closing note
Return to Luo Hui's industrial analogy.
The industrialization of physical products took over a century: first electricity, then machine tools, build the assembly line, finally build lights-out factories. The industrialization of brainwork has only just reached step two. GPT is electricity; various models and Agents are the machine tools; the assembly line is being built; the lights-out factory is far away.
But the speed is totally different. Physical-product industrialization took generations. Brainwork's may happen within these one or two years.
This assembly line won't wait for anyone. It doesn't care whether you believe in AI; it only cares whether you're tightening screws.
Those already moving — the ones doing tens of millions in revenue with 20 people, replacing an entire new-media department with AI, generating stories for kids globally with foundation models — they aren't betting on the future. They're defining the standard.
If you wait until the assembly line is built to enter, you may not even qualify to tighten screws.
What do you think?
More Conversation Detail
Panelists:
Qiyu Shuke Founder — Shen Yang (沈扬)
Ji Su Engine Founder — Wang Jun (王俊)
Guomai Zhiyu & Xingju Tong OpenReels Co-Founder — Li Deheng (李德恒, Henry)
Yunxigu Technology Founder & CEO — Luo Hui (罗辉)
Host: Unique Capital VP — Wu Shenliang (吴申亮, Jeffrey)
I. Panelist self-introductions and years using AI
Jeffrey: Today's theme is "The infinite studio: how AI industrializes global content." Over the past year, AI tech tied to content production has advanced extremely fast; overseas video models and Douyin's advanced models aren't just changing the production end, they're restructuring content structure. Our four guests are all industry veterans; first let's have each briefly introduce the company business and how long they've guided services with AI. Start with Mr. Shen.
Shen Yang: We use AI Agents to do new-media marketing for vertical industries (mainly autos now). The core is helping auto-dealer 4S stores do short-video production, livestreaming, ad placement, operations, and host/customer service. We use digital employees to replace humans, helping merchants deliver the full new-media chain with AI.
Our core logic is "effect-based delivery": the client does nothing; after initializing devices, going live, video production, ad placement need no involvement; we directly deliver sales leads. Currently most of China's auto OEMs and over 60% of dealer groups use our product; we've achieved some results in the vertical.
Luo Hui: Ours is a bit different; Yunxigu is a children's-education smart-hardware company. Because it's children's education, it needs lots of content. Internally we often say the hardware product is just "bowls and chopsticks," while the content is the "meat and greens." We provide tools to let kids absorb educational content better. In these years of the foundation-model wave, we're heavily using AI to industrially build educational content.
Wang Jun: Ji Su Engine is a company integrating AI, IP, and digital assets, mainly providing AI solutions for IP incubation and operations, like the popular AI short dramas and AI Agents. Baidu, iFlytek, and Google are all great partners. We were one of last year's Google Going Global Accelerator companies, and the only AI content-operations one. Two days ago we just represented Nanjing at the China Cultural Industries Fair, displaying in the Nanjing section at the International Expo Center through the 25th; interested readers can go see.
Li Deheng: Guomai is a new publishing company dual-driven by AI and IP. Over the past decade-plus we've accumulated book-publishing experience, signed outstanding authors like Yi Zhongtian, Han Han, and Luo Xiang, and done a lot of film/TV exploration. Entering the AI era, Guomai Zhiyu will, based on installed IP and original strategy, do further AI-visualized content creation, IP adaptation, global distribution, and full-chain IP operations. We're still newcomers in the AI industry, coming with a learning mindset.
II. The AI content ecosystem: a tool, or a standard industrial system?
Jeffrey: People often feel AI content production is just a tool for creating images, text, or video. But Qiyu serves so many auto B-side clients; Mr. Shen, do you think AI will eventually become a standard, automated industrial system in the whole content ecosystem?
Shen Yang: The future direction is definitely this. Traditional businesses (like auto 4S stores, home-goods stores, real-estate agencies) face the pain of moving from offline acquisition to online, because mature hosts and shooter/editors are hard to hire and keep.
For traditional industries, AI in the future isn't just a tool; it is the new-media department's employee. Practice shows that as long as you fully AI-ify and Agent-ify the workflow, video production and livestreaming can be fully automated. But how to avoid content homogeneity and keep up with platform algorithm trends to produce hits needs human judgment and coordination. The production process, though, has great hope of becoming factory-like.
Jeffrey: Video models are developing extremely fast this year. But there's a real problem: between "can generate" and "can commercially use" there's still a big gap; premium content still needs human intervention. Mr. Wang, after Ji Su Engine's real experience, how far is AI from fully automated production?
Wang Jun: First, our company is now close to semi-automation, all-AI, no step without AI.
On whether people are needed, we must clarify a concept: market AI companies split into two types. The first treats this as an e-commerce thing, not doing content; don't consider them. We're discussing the second type — companies that truly do content.
People talk about content consistency and control, but I'd say: from the earliest AI training paradigm, it was fated that full automation isn't easy now. Why? Because AI training's underlying premise is "tell AI humans like this, and it produces similar things for a high score." But "what's right and what we need" differs from "what humans like"; what AI produces is often an average. These two are fundamentally at odds.
Pure automation isn't possible short-term because the training paradigm hasn't changed. But in the future, optimizing on top of Agents — for example calling Skills via MCP — can optimize it on another level. That day will definitely come.
Jeffrey: Next question to Mr. Li. Guomai Zhiyu holds lots of IP resources and has invested heavily in overseas expansion. With AI now fully empowering, what new opportunities are there for IP creation and overseas expansion?
Li Deheng: We aren't creating generic entertainment content from zero to one; we take the classic IP in hand for global distribution and operations. Our core moat and global confidence lie in "how to tell the China story well."
AI has changed past creation/production logic enormously. Before, adapting a literary IP to film/TV was an extremely long chain — script, characters, plot, shot control — consuming huge director-team effort, and couldn't concurrently do multiple narrative adaptations; cost was too high. After AI, you can make standardized derived multiple creations around the original's core narrative thinking, a qualitative change in production efficiency.
From the IP-going-global angle, the old barrier wasn't just language; much of it was narrative logic and aesthetic preference. In the past you had to do overseas localization and overseas shooting; over-adapting often killed the IP's original vitality. Now, based on a unified narrative background, AI can provide extensible localization space while keeping standardized creation.
Jeffrey: Last, separately ask Mr. Luo; here only Yunxigu does "hardware + AI." A more foundational question: today's biggest limit on the AI content industry is model capability, or the upstream supply-chain ecosystem? Where's the biggest gap?
Luo Hui: The conclusion: AI content's whole industrialization has only just begun; the big wave is still early. It isn't fully landed yet mainly because the industrialization process is immature; the whole ecosystem needs time to mature.
Yunxigu's whole company is very AI-ified, while shipping 10 million devices a year. We call hardware "physical products" and AI content "brainwork products." We can compare the two products' industrialization:
Physical products: first "electricity" (energy), then lots of "machine tools" (surface-mount, wave soldering, etc.); with electricity and machine tools, start building the "assembly line," later the "lights-out factory."
Brainwork products: a few years ago GPT-3.5 came out; foundation models appeared like "electricity." These two years of massive Agent apps, whether OpenClaw or Claude, are the "machine tools" of brainwork products.
Now the "machine tools" have just started maturing, so we can talk about the assembly line. Our goal these two years is very clear: from a development company in an office building to one striving to build a brainwork-product assembly line. Right now brainwork "machine tools" produce relatively poor consistency, still needing human re-inspection. Whoever builds this assembly line most perfectly leads. As for the "lights-out factory," we're still exploring; we've just reached the moment of "having electricity and machine tools."
Also on division of labor, phone production may take hundreds of companies; brainwork products now are mostly text-to-film directly, but future ecosystem division will also emerge. Everyone is still figuring out how to build the assembly line and use fewer people for higher efficiency.
III. What is the industry's scarcest resource?
Jeffrey: AI has exploded content supply, but user attention hasn't increased. After content production cost drops sharply, what is the industry's scarcest resource? IP, channels, or platforms?
Li Deheng: From the PC era to mobile internet, from image/text to short video, information keeps exploding. The scarcest capabilities are two:
The ability to keep producing quality, original-worldview products — the true content-creation source with thought and value.
The ability to precisely grasp emotional resonance across demographics and occupy user mindshare. The content era will also produce content brands; that's the most critical.
Luo Hui: I'll use "eating" again. In the past when the economy was bad, everyone was underfed and happy with steamed buns. Now everyone's full; after content cost drops, how do you make consumers like eating yours?
Scarce ingredients and environment: you hold certain special digital content, producing "food" only you have.
Personalized service: the moment a customer walks in you know whether they eat spicy; more grasp of users' personalized data. Getting ingredients, environment, and service right is most important.
Jeffrey: Interrupt — Mr. Shen and Mr. Wang's companies are raising funds now; investors most love asking: models get more open-source and homogeneous, in the end, what's your real difference and moat from other companies?
Shen Yang: Following Mr. Luo, everyone's core moat in the future isn't productivity but "the feel for beauty and aesthetic sense." Good content brings aesthetic and spiritual enjoyment. In the end it's the team's cognition of beauty, and the ability to industrialize the production of beauty.
Some beauty is art, but some beauty (like the first-generation iPhone), once defined, can become standardized production. That needs extremely strong aesthetic understanding and engineering breakdown. The future competitive barrier is whether a team combines aesthetic judgment with technology — able to judge good content and industrialize production.
Jeffrey: Sounds like future prospects for art students may beat science students. What does Mr. Wang think?
Wang Jun: Aesthetics is undeniably important. But think differently: future content forms may be disrupted. Today's AI production is still the traditional model (how many minutes per episode), but next it'll hit a bottleneck — is Douyin's traffic pool suited to display massive content?
Future content forms may be "per-person." 100 people here watching the same drama, underneath it's one IP but the content is totally different. I like the cute type, he likes the elegant type, the shown content fully fits individual preference.
As for the moat, we make Agent tools; wanting to generate 80-point quality with one sentence, no threshold, everyone is similar now. The real moat is the team itself. We aren't afraid of poaching, because poaching a director from me to another may not fit. We're a Team; internal org structure, role naming are totally different from traditional film companies; the production method is fully innovative.
IV. Does the next-generation content company still have geographic identity?
Jeffrey: In the old film era, going overseas had cultural and production-system barriers. Now dubbing and subtitles can be fully replaced by AI; a "global content factory" seems within reach. Will next-generation content companies still have obvious national or regional identity? Mr. Li, your globalization experience?
Li Deheng: In channel distribution and AI-tool usage, globalization and automation are fully achievable. But the carrier of story and content itself can't fully shed national or cultural identity.
The same work, in different regions, needs adaptation to local narrative modes and aesthetic structures. A creator's life experience can't spread globally without difference, so the story core still has a clear local stamp. But AI can do standardized extension and modification based on the original IP story core. The end picture may be: technology and channels global, but local story cores very regional.
Jeffrey: Which global market has Guomai done best so far?
Li Deheng: We just took the first step; currently we take Hong Kong as a stronghold, trying to distribute historical or pop-culture themes, applying the distribution-channel methods once common in e-commerce.
Wang Jun: Speaking of this, I'm especially excited. Before, shooting a live-action drama, getting on the domestic table cost 1.5 million RMB, and going overseas for distribution cost another $200,000 to $300,000 to reshoot. Now it's totally fine; all the minor languages we didn't dare touch are doable. But on globalization, production is definitely global, the consumption side is definitely local. Each country's IP development and consumption logic is totally different.
Jeffrey: Mr. Shen, any thoughts?
Shen Yang: Cross-border content creation is similar to cultural differences across China's different regions and provinces, just amplified. In the AI era we can push trial-and-error cost extremely low.
Before, content marketing was "white-box," relying on people to deconstruct hit logic; I think in the AI era we can do "black-box." Algorithm platforms give clear data feedback on exposure-to-conversion; we can use AI to build workflows, run low-cost massive-testing black-box evolution, and output content that truly fits the locale. The parameters that worked for us in Sichuan-Chongqing, the Northeast, and East China are completely different; we fully follow A/B test results. In the future it's aesthetics, understanding of algorithm platforms, and engineering ability.
Jeffrey: Mr. Luo, what do you think?
Luo Hui: Will local companies achieve globalized content industrialization? We think 100%. After physical-product industrialization, lots of multinationals spread worldwide. Today much AI content already crosses borders; although there's 20% special local content, 80% is global.
Our children's stories already sell in volume to countries worldwide. We regenerate content by foundation models based on different countries' cultures; practice proves cultural differences can absolutely be replaced by industrialized technology.
V. One-year outlook for the AI content industry
Jeffrey: Time's about up; last open question. A year from now, what changes in the AI content industry? (New professions, new business models, or new platforms.) Please summarize.
Luo Hui: Writing software to make content used to cost a million; in the future 100 RMB may do it. After content cost drops to extremely low, business models will change hugely. Buying a product used to be a gamble; in the future product sales may basically become free, shifting to "post-pay by effect" or "use-and-take."
Wang Jun: Change is too fast; maybe in half a year it's totally different. Today's tools all talk about efficiency, but this is far from the final form. Including distribution — could there be a tool where content is already distributed and being watched the moment it's done? Maybe there's no "distribution" step. As for role changes, our company internally isn't traditional film-company roles anymore; it's already happening.
Li Deheng: Future digital-cultural content may evolve into an asset-like form, able to be rights-confirmed and traded. The AI field has spawned many new roles, new assets, new plots; this will spawn new platform companies that provide foundational capability in digital-asset rights confirmation and IP-asset trading flow.
Shen Yang: An in-vogue phrase at our company: teams that finish AI-ified coding first will laugh at teams not using AI for "old-way coding."
Development cost dropping sharply, plus AI iterating weekly, next year content is definitely a big-bang era. On the marketing end, every salesperson or store will have an AI-built digital IP. It's like "professional beauty filters" — most people can't bear the workload and mental pressure of posting daily and livestreaming, but AI can turn a living person into the most perfect digital IP, helping you present the product better. This is our vision.