Original · Unique Research · 2026-03-18
Editor's note: This is a complete historical interview account and accompanying commentary. First-person assessments and praise belong to the original author; product positioning, delivery volumes and strategic intentions are attributed to the source or founder rather than independently tested findings. Kuangye Qunxing, Liu Rushan and Jianying are romanizations of the source names, not assertions of official English legal names; MulanAI is the source's product spelling. The reported monthly delivery of more than 600 videos and peaks above 1000 have not been independently audited, and do not establish revenue, customer outcomes or comparative superiority. “White-box,” “Agent First,” “operating system” and infrastructure describe the company's approach and ambitions, not certifications or verified interoperability. Predictions about creative work, art pricing, industry structure and the end of 2026 remain historical opinions. The source's caveat that market adoption may be slower than expected is retained, as are all twelve Q&A pairs; no product guarantee or current recommendation is implied.
Unique Awards · Guest Interview
Why Does This AI Video Company
Insist on Braking While Everyone Else Sprints?
Liu Rushan, founder of Kuangye Qunxing: The real danger is not moving slowly, but appearing to move fast.
Over the past two years, speed has been the one thing the AI startup world has not lacked.
Connect a model today, launch a feature tomorrow; talk about Agents this week and workflows the next. Someone is always posting a new version, a funding announcement or a growth curve, as though running fast enough lets you claim the future first.
The problem is that much of the speed in AI is not genuine speed.
It is closer to performative acceleration: products appear to update frequently, features multiply and the story becomes more seductive. But put them into real business operations, under delivery pressure, where customers need hundreds of pieces of content a day, repeated revisions and dependable batch production, and the shortcomings quickly show.
After a recent conversation with Kuangye Qunxing founder and CEO Liu Rushan, one sentence stayed in my mind:
"
The real danger is not moving slowly, but appearing to move fast.
In an industry atmosphere of building first and asking questions later, Liu Rushan and her team did something counterintuitive. They did not rush to trade a half-finished product for attention or package themselves as another fast, cool AI video product. Starting in 2024, they took the R&D team directly into real commercial video work, building AI video-marketing portfolios for listed companies and deliberately placing themselves on the production floor.
They delivered more than 600 videos a month, exceeding 1000 at peak periods.
You could see that as braking.
But look more closely and they were not braking R&D. They were putting a brake on the industry's impatience.
It was not that they did not want to move fast.
They wanted first to understand whether AI video meant building a smarter toy or a system that could genuinely enter a commercial production line.
That question may be more important than which model comes next.
Not Making One Video, but Rebuilding a Production Line
There are already many AI video products on the market.
Some generate a video from a sentence, edit one with a click, write scripts or create storyboards. It all looks lively, and first-time users can genuinely feel that a new era has arrived.
But take one step further into business use and the problem turns out to be far less simple.
The commercial world has never needed something that impresses only occasionally.
It needs to know whether you can deliver today, tomorrow and after ten rounds of revisions. Can you handle a customer asking to keep this shot, preserve that action, change the clothing, rewrite the subtitles and strengthen the branding? Can you remain dependable when a team starts testing batches across versions, channels, languages and audiences?
This is where the limits of many AI video products become visible.
They are good at generation, but not production.
They can produce an output, but cannot support a process.
They resemble creative assistants more than production systems.
Liu Rushan's definition of Kuangye Qunxing is worth remembering. She says they are not building an ordinary AI video tool, but video-production infrastructure for the AI era.
That sounds grand, but in plain language it means one thing:
They do not simply want to help you make one video. They want people and AI together to run the entire chain from idea to finished video, and from individual pieces to batches.
MulanAI's core is therefore not one spectacular feature, but the combination of several things:
A canvas, workflows, compositing and editing, and Agents.
None of these terms is new on its own. The difficulty is turning them into one system.
Once you start thinking in systems, the product logic changes completely.
Much previous software assumed its user was a person.
People click buttons, look at interfaces, respond to interaction design and switch between tools themselves, filling gaps and making judgments.
The Agent era is different.
Increasingly, tasks are being completed not only by people but by Agents.
An Agent will not be persuaded by a prettier homepage or fall in love with a product because its marketing video looks cooler. It cares about whether the system can be understood, called, combined and reused to complete a task reliably.
That is why Liu Rushan says they have been working toward one idea from very early on:
Make something agents want.
Put plainly, while many people still design video software around making button-clicking easier for humans, they are rewriting it around how Agents understand, execute and collaborate.
This is not a UI upgrade.
It is a paradigm shift.
OpenClaw's Popularity Made Their Approach Easier for Me to Understand
Many people have discussed OpenClaw over the past few months.
Different people explain its popularity differently: some see the product form, some the community momentum, and others the boundary between AI that talks and AI that acts.
But viewed in the broader evolution of software, OpenClaw has brought a more fundamental understanding into focus:
The next generation of software is not merely a smarter chat window, but an entity that can genuinely keep executing work.
This has major implications for many industries.
The effect on video is particularly significant.
Video production involves long processes, extensive collaboration and frequent revisions. It is inherently ill-suited to an approach based only on generating an output inside a black box.
Customers will not be satisfied with simply asking to see a finished video.
More often, they say:
This shot is good. Leave it alone.
Replace this character.
Keep the composition.
Rewrite the copy.
Make the pacing faster.
Send this version to TikTok and that one to YouTube Shorts.
Make five more variations of this material and get them to me tonight.
This is not fundamentally a question of whether something can be generated.
It is a question of whether production can be supported.
If every revision forces a system to start the whole video again, even a very intelligent system will struggle to enter real business workflows.
Liu Rushan's team therefore insists on a white-box approach, workflows, a free-form canvas and Agents within the system. The point is not to appear technically advanced, but to answer the same question:
How can AI genuinely enter production rather than remain a performance?
I see this as one of the biggest differences between MulanAI and many other AI video products.
Others may be using AI to improve the previous generation of video tools.
They seem instead to be rebuilding a video-production system for the AI era.
The two paths may not look very different at first. The latter can even seem slower, heavier and less attractive.
But as the industry moves from experimentation to production, the dividing line will become clearer.
The former optimizes a single experience. The latter optimizes long-term production capacity.
The former competes to become popular more easily. The latter competes to survive more easily.
The Real Moat Has Never Been a Few More Features
Discussions of AI startups easily arrive at a familiar question: what is your barrier to competition?
Many answer with models, algorithms, team backgrounds, industry resources or distribution capabilities, or simply offer impressive-sounding technical terms.
Liu Rushan's answer seems closer to reality.
She says Kuangye Qunxing's hardest-to-copy asset is not an individual technology, a particular industry resource or merely an overseas distribution channel. It is a systems capability.
That is right.
The easiest things to copy today are precisely the superficial ones.
A feature, a page, a model integration, an impressive demo or even a product narrative can quickly be matched. Build it today and someone else can catch up tomorrow.
The difficult question is whether you genuinely understand:
How video production should be rebuilt in the AI era.
That understanding comes from at least three layers.
The first is technical judgment.
Did you recognize early that connecting a few models to make videos would not be enough, and that products must be designed Agent First, organized around workflows and built as systems rather than feature collections?
The second is industry understanding gained through actual production.
Have you been on the production floor? Experienced genuine delivery pressure? Been cornered by customer requirements? Been taught hard lessons repeatedly in situations where something must be delivered today?
The third is the ability to turn the first two into a product.
Many people know the problems and many understand technology. Few can combine judgment about the direction, industry understanding and user experience in one system.
Ultimately, the barrier is not what you know.
It is what you have built into something other people cannot do without.
That is why Liu Rushan says they do not want to be a feature company or a UI layer wrapped around a model. They want to build a video-production operating system.
Not because OS sounds more attractive.
On the contrary, it is a distinctly unglamorous path.
It means dealing with complexity, collaboration, revisions, batch production, organizational use, model switching and the constant tension between quality and efficiency.
But precisely because it is difficult, successfully following that path puts you in a different position.
Features can be absorbed into other products.
Systems are harder to absorb.
AI Is Replacing Inefficiency in Video Production, Not Video Creation Itself
When discussing AI video, people most often ask:
Will it replace video professionals?
I increasingly think the question itself is flawed.
What AI replaces is not video creation as a whole, but the repetitive, mechanical, fragmented execution work that depends heavily on manual handoffs.
First-draft scripts, shot breakdowns, text-to-image, image-to-video, voiceovers, subtitles and basic finishing once required coordination among different roles. Those stages are now being compressed rapidly.
Teams also used to switch repeatedly between models, software packages and tools, creating a fragmented chain with high communication and experimentation costs. The value of workflows is fundamentally to connect those scattered capabilities into an executable sequence.
Another area being noticeably compressed is inefficient trial and error.
Making ten versions and testing different assets and narrative structures used to take substantial time and money. AI has lowered those barriers significantly.
Interestingly, it is often execution that is replaced while judgment is amplified.
Strategy becomes more valuable.
Creativity becomes more valuable.
Aesthetic judgment becomes more valuable.
Systems operations become more valuable.
So does the ability to orchestrate Agents.
As making something becomes easier, knowing what is worth watching becomes scarcer. When generation is no longer scarce, turning generation into production becomes scarce.
That is why I think the real differences in AI video will depend not merely on model capability but on who can respond to this redistribution of value.
Execution is automated, judgment is amplified and systems are revalued.
MulanAI clearly does not want to occupy the first of those positions.
AI Lets Ordinary People Make Videos, but That Does Not Make Art Less Valuable
I particularly liked Liu Rushan's view of artistic barriers in this interview.
She says AI is indeed lowering barriers, but mainly barriers to production, not to artistry.
I strongly agree.
Many people previously could not make things not because they lacked feelings or ideas, but because they did not know the complicated software, processes and techniques. AI gives more ordinary people their first opportunity to express their ideas, which is certainly progress.
But the quality a work can reach has never ultimately depended on mastery of a particular software package.
It depends on a distinctive perspective, authentic expression, aesthetic judgment and the ability to move people.
When everyone can create reasonably good content, those qualities become scarce again.
AI has not made art disappear.
It has merely started to undermine the artificial scarcity that previously came from something being difficult to produce.
What will lose value is not art, but the pricing structure for mediocre content.
This may make the market more honest.
Once technology is no longer a barrier, only two genuinely valuable things remain:
Results-oriented systems for efficiency.
And irreplaceable expression.
The former belongs to industry; the latter to art.
Both paths will become clearer.
This Company Wants to Execute Better, Not Merely Generate Videos Better
If I had to sum up my strongest impression of Kuangye Qunxing and MulanAI in one sentence, it would be this:
They are not seeking a place in a burst of AI video excitement. They are betting on the right to define the next generation of video software.
There is a substantial difference.
Competing for attention emphasizes growth curves, reach and market visibility.
Competing to define the category emphasizes understanding what the industry will become.
Liu Rushan's view is clear:
Future video software will not merely generate better but execute better. It will be a production system, not just a creative tool; infrastructure that Agents can understand, call and reuse, not merely an interface for people.
If that judgment is right, many impressive-looking individual capabilities today will ultimately be transitional.
What will be truly valuable is the system that supports complex processes, commercial delivery, organizational collaboration and production at scale.
Of course, that path may not be the easiest story to tell or the easiest for people to understand immediately.
Even by the end of 2026, it may face a risk: the market may not shift from flashy individual features to systems-level production tools as quickly as the team expects.
But as Liu Rushan says, if large technology companies turn AI-powered international content into a fiercely competitive market, they will not compete with those companies over entry points, standard offerings or generic features.
They will move into a deeper layer and tackle what large companies find hardest to develop deeply:
Vertical-industry content-production systems, complex workflows and deliverable results.
I remembered this line too:
They sell results, not capabilities.
Many companies build standard offerings.
They want to build something customers depend on.
That may be the most valuable position an AI company can occupy.
Closing Thoughts
What has made AI so intoxicating over the past two years is the feeling that everything is happening at once.
New models every day, new products every week and new concepts every month. It is easy to be swept along, feeling that slowing down at all means missing every benefit of the era.
But Liu Rushan and Kuangye Qunxing's approach offers an important reminder:
Not every kind of speed is worth pursuing.
Some kinds lead to bubbles.
Some speed means pushing unfinished products into the market faster.
Some growth is merely a temporary illusion before the market has finished learning about the category.
What ultimately endures through cycles is what appears slow but is actually deep.
Understanding real production environments, for example.
Reconstructing workflows.
Or turning AI from a feature that generates into a system that executes.
So if you still see MulanAI simply as an AI video company, you are underestimating it.
What it seeks may not be a position as a tool.
It may be a position as a system in the AI-era video-production line.
That is not a more attention-grabbing position.
But it may well be a more enduring one.
Selected Q&A
1. Why did Kuangye Qunxing choose to brake when AI startups emphasize speed so heavily?
They are not braking R&D but the industry's impatience. Liu Rushan sees the real danger not as slowness but false speed. Many products seem to update quickly and offer many features, but without entering real production environments and facing delivery pressure, the result may be a clever-looking demo rather than a product that can enter industry workflows.
2. What is the biggest difference between MulanAI and ordinary AI video tools?
In a sentence: others mainly build AI creative tools, while MulanAI wants to build an AI-era video-production system. Many products ask how to make it easier for people to create videos. MulanAI asks how to give AI genuine video-production capabilities and let people direct that system. It emphasizes the integration of canvas, workflows, compositing and editing, and Agents. The goal is not single-video generation but an entire executable, reusable, scalable production chain.
3. What did OpenClaw's popularity teach MulanAI?
Liu Rushan's view is direct: OpenClaw helped more people understand for the first time that AI does not just talk; it is beginning to act. That reinforced their belief that future software's default users will include Agents as well as people. Software must therefore be designed not just around how people click buttons, but around how Agents understand, call and execute. She noted that the team had long maintained an internal principle: Make something agents want.
4. Why insist on a white-box approach instead of black-box, one-click video generation?
Commercial video production is not as simple as producing one output. It involves constant revision, collaboration and delivery. Customers ask not only whether a video can be generated but whether they can keep the framework while changing details, run batches, collaborate as a team, avoid returning to Jianying for manual compositing and truly integrate the work into business processes. A black box can provide an output; a white box can support revision, collaboration and delivery.
5. What is Kuangye Qunxing's hardest-to-copy capability?
Not an individual technology or resource, but a systems capability combining three layers: forward-looking judgment about AI video's future production methods; industry understanding refined through real commercial delivery; and the ability to turn those two into a systematic product. What is hard to copy is never superficial functionality, but whether understanding the problem has become something others cannot do without.
6. Who are MulanAI's core users today?
Not occasional users trying something new, but people who make videos frequently and continuously for commercial purposes: professional cross-border marketing teams, brand-content teams, agency operations teams and professional commercial creators. They share a concern not with whether AI is entertaining, but whether videos can be produced reliably and at scale and deliver actual business results.
7. Which stages of video creation do AI workflows replace?
Mainly the repetitive, mechanical and inefficient parts of execution: first-draft scripts, shot breakdowns, text-to-image, image-to-video, voiceovers, subtitles, basic finishing, friction from switching tools, inefficient experimentation and some mechanical work in junior execution roles. At the same time, strategy, creativity, aesthetics, systems operations and Agent orchestration become more important.
8. Will AI devalue art?
Liu Rushan believes AI lowers production barriers, not artistic ones. What loses value is not art, but content valued for being difficult to make rather than distinctive. AI will undermine mediocre content's pricing structure while allowing truly scarce expression, aesthetics and perspectives to re-emerge. In other words, it does not make art cheaper; it makes the art market more honest.
9. Will creators become prompt engineers in the AI era?
Liu Rushan sees prompt engineer as more of a transitional concept. What determines a work's quality is not prompting technique but the creator's agency: are you using AI, or being led by its default, average aesthetic? Are you expressing yourself, or consuming the model's hallucinations? Weak creators may become prompt engineers, but genuine creators remain artists.
10. What growth opportunity does Kuangye Qunxing see most strongly over the next 12 months?
Infrastructure for scaling professional content. AI video's first wave solved the move from 0 to 1, allowing anyone to make videos. The next opportunity is going from 1 to 100: helping enterprises, brands, MCNs and professional teams produce large volumes of efficient, multiversion, personalized videos while maintaining quality and brand identity. This is an upgrade not only of tools but of how production is organized.
11. What if large companies make AI-powered international content a fiercely competitive market?
Liu Rushan's answer is clear: do not compete over entry points, standard offerings or generic features. Move deeper, into what large companies struggle to develop fully: vertical-industry content-production systems, complex workflows and deliverable outcomes. She puts it plainly: large companies can take traffic and foundational capabilities, but not the organizational coordination, accumulated data and end-to-end business processes within customers' daily production. Others sell capabilities; they want to sell results.
12. What kind of company does Kuangye Qunxing want to become by the end of 2026?
Liu Rushan expects it to evolve from a company making AI video tools into an AI-native video-infrastructure company for global commercial content production, enabling brands, teams and creators not merely to generate videos but to produce them continuously, controllably and at scale.
This document is original content by Unique Research.