Original · Unique Research · 2025-11-13
Editorial note: This is a complete English edition of the historical article and interview. The launch comparison with Microsoft Copilot, registered-user and service-user totals, financing, model counts, technology capabilities and partnerships are claims reported by the source or interviewee, not independently verified current results. The source does not identify which Copilot announcement or launch underlies the 12-day comparison. Its references to roughly 13 years of experience and the 2015–2025 product timeline are retained as written; “this year” and other relative timing remain anchored to the historical source. BiYou Technology is a romanization of 必优科技, not a separately verified official English corporate name.
On many office workers’ lists of nightmares, PowerPoint surely ranks among the top three. The title needs logic, the content needs data, the layout needs aesthetic appeal, and in the end you still have to present it yourself. As tools grow ever more powerful, many people feel they are becoming less and less able to make PPTs.
But there are always people who move in the opposite direction from the public’s pain points.
While others are figuring out how to escape PPT, Zhou Ze'an chose to spend a lifetime grappling with documents.
He has spent nearly 13 years on the AI+ document track, a niche as narrow as they come.
He built AI document products, with a venture later acquired; led the incubation of WPS Smart Documents at Kingsoft; then struck out again in 2020—still building AI document products.
Many see this as an obsession; he himself prefers to call it this: returning to the essence of document creation.
I. Why Would Anyone Spend 13 Years Competing in Documents?
Judging from his résumé alone, Zhou Ze'an is a typical serial entrepreneur:
In 2015, he built an AI document product, proved out its commercial viability, and the venture was acquired; later, he participated in WPS Smart Documents at Kingsoft Office; in 2020, he founded BiYou Technology in another startup venture.
But if you lay out the timeline, you find that he has been doing the same thing all along: using AI to turn documents back from a tool burden into containers for content.
From the outset, BiYou Technology has been a vertical-application AI company.
Its earliest work was in specialized document models: AI résumé writing, AI contract writing, and other highly specific scenarios—where you can see the problem, quantify the outcome, and clearly know whether users are willing to pay for it.
These seemingly fragmented scenarios allowed the team to accumulate two very important things:
The complex details of real-world document scenarios, rather than perfect samples from a laboratory
A complete closed-loop body of experience from models to products to commercialization
When the large-model wave arrived in 2023, many people hastily changed course and began attaching AI labels to themselves.
What BiYou Technology did was launch the experience it had accumulated over the years in a new form—ChatPPT.
This product launched 12 days before Microsoft Copilot, added one million organic native users in 45 days—and single-handedly turned AI-powered PPT creation into an independent category.
Today, ChatPPT has a personal edition, enterprise edition, platform edition, AIPC edition, AI-glasses edition, and the overseas edition Utalk, with more than 15 million registered users.
If you look only at these numbers, it is easy to see it as another breakout AI tool.
But in Zhou Ze'an’s mind, it is more like a prototype: a personal super-document office workstation for the AI era.
PPT is only the entry point, not the destination.
II. PPT’s Four Hidden Pitfalls—and ChatPPT’s Four Answers
To understand what ChatPPT is doing, it is better to return first to a scenario every office worker knows well.
1. The More Features There Are, the Less You Know How to Use Them
Traditional Office software has no fewer than 1,800 features to start with.
Every button is useful, but layered together they become a maze with an extremely high barrier to entry.
Many people’s real mindset when making a PPT is:
I’m not expressing a point of view; I’m fighting the tool.
ChatPPT’s first move is to cut out that fighting stage.
It insists on doing something that sounds somewhat extreme: supporting creation only through conversation.
You tell it what you need: Help me make a pitch deck for a presentation to the CEO, no more than 10 slides, focusing on the product’s profit model.
Let the AI parse the instructions, adjust the formatting, and choose the template.
Users are responsible only for the outcome, not for the process.
Zhou Ze'an has coined a term for it: LGUI (Language Guide User Interface)—using language to drive the interface.
Put simply, it hides complex tools in the background and turns a single sentence into a true super-command.
2. If the Content Is Made Up, Who Would Dare Show It to Their Boss?
The most unsettling thing about generative AI is that it can spout nonsense with a straight face.
In a chat, you can laugh it off; when pitching onstage or reporting to management, no one dares take that gamble.
Office documents are things for which you must be accountable for the results—getting one number wrong can cause trouble.
So BiYou Technology did something that looks very conservative:
It added a layer of traceable-source generation to the generation process.
Whether it is pulling data from webpages or uploading your own documents, every time ChatPPT generates the content of a slide, it can tell you:
Which passage and which Q&A exchange the point on this slide came from.
You no longer have to simply trust it; you can verify it.
AI is no longer a black box, but an assistant that lays out its reasoning process.
3. The Content Is Good, but the Layout Is So Ugly It Hurts the Presentation
Making a PPT involves two hurdles:
The first is content logic; the second is visual presentation.
Many people are already exhausted by the first hurdle, only to fall into the hell of aligning objects in the second.
Zhou Ze'an simply handed the second hurdle over to AI as well:
The team developed its own Layouter model, dedicated specifically to layout and visual polishing.
It can do several things:
Whether it is a PPT you made yourself or one generated by AI, it can be reformatted with a single click.
You can upload style templates you commonly use and let the model learn to lay things out the way you do.
The result is:
PPT is no longer a question of whether you have time to do the design; it becomes a question of which style you want.
4. The Last Mile: It Is Not About Finishing the PPT, but Presenting It Well
PPT has one feature that differs from both Word and spreadsheets:
It was born to be presented, not merely viewed.
Many people’s actual state before taking the stage is this:
The script is not finished, the animations have not been adjusted, and there are only four words in their heads—“It’s over, it’s over.”
To solve this last-mile problem, BiYou Technology did three more things:
It automatically generates a speech script, helping you translate the logic in your PPT into language you can actually say aloud.
It automatically creates interactive animations, so the pacing of the slides matches your presentation.
It developed an AI roadshow feature and even partnered with Rokid and RayNeo so that the speech script can be projected into AI glasses, following you like a teleprompter.
You look at the audience, and the audience looks at you.
And you discreetly look at the script and page numbers in your glasses.
The technology is working furiously behind the scenes, while the presenter onstage appears composed and natural.
At bottom, what ChatPPT does is very simple:
It frees people from the details of tools, allowing their energy to return to what they want to say.
III. From the “Toolbar” to the “Dialogue Box”: What Does This New Species of Document Look Like?
If traditional Office is understood as a big general store, then what BiYou Technology is doing is breaking it apart into little intelligent companions that can each do work.
Technically, they are pursuing a large-model-plus-small-model approach:
At the base layer, general-purpose large models ensure comprehension and language capabilities;
On top of that, they stack 28 small models tailored to document scenarios, ranging from 100 million to 2 billion parameters.
You can think of it as a generalist brain accompanied by a group of specialist teaching assistants.
More interestingly, they are pushing the entire document system in the direction of Doc-AI-Agent:
Documents are no longer passive files, but living objects.
They can carry out Actions according to your instructions, call tools, and connect with other applications.
For example, they embraced Agent protocols such as MCP at a very early stage.
If AI applications are compared to a pile of electrical appliances, MCP is like a unified plug standard.
Once connected, they can call on one another’s capabilities and pass information to one another.
One step further is LiveDoc:
A living document format redesigned for the AI era.
In this vision, a document is no longer a file that stores a pile of static text and charts,
but a work interface that can be continuously updated and continuously linked with other systems.
You no longer need to distinguish whether this is a PPT, a Word document, or some other format;
You only need to care about: what I want to express, who it is for, and how it will be used.
IV. Technology Should Not Be for Show; It Should Be a Craft That Can Make Money
Many AI startups share a common dilemma:
The technology team wants to be cooler, while the business team wants to be steadier.
Zhou Ze'an’s attitude is very straightforward:
Innovation should come first, but innovation that only looks good and does not make money is meaningless.
He has one small addition to make about innovation:
It does not necessarily have to be a major breakthrough in underlying algorithms;
Redesigning products and processes within user scenarios is also a form of technological innovation.
For example, using LGUI to replace traditional menu-based interaction,
using traceable generation to turn fabrication into verifiable generation,
or using the Layouter model to turn whether one can design into a switch.
These may not look like paper titles, but they are what users genuinely use every day.
At the same time, he places enormous emphasis on the other half: commercial validation.
In his view, there are three fundamental principles:
Innovation must come first; otherwise, in the AI era, you are simply not qualified to sit at the table.
Every innovation must quickly find a setting where it can be validated by real business.
The relative weight of technology and business can be adjusted dynamically, but neither can go it alone.
You can see BiYou Technology's rhythm:
It first built a foundation through vertical document use cases, then used ChatPPT to drive growth, and then entered the B2B market through its enterprise edition, API, SaaS, private deployment, and other approaches;
At the same time, it secured two rounds of financing from Kingsoft Office and Baidu, giving this path not just a story but cash flow as well.
In his view, a truly good AI product lies at the intersection of what technologists find interesting, users find easy to use, and financial statements find reasonable.
V. Sharing a Table with the Giants
When it comes to office documents, several giants are impossible to avoid: Microsoft, Kingsoft, and even various local Office solutions.
Many people ask: isn't starting a business in this track inherently a disadvantage?
Zhou Ze'an's answer is, on the contrary, quite optimistic.
On the one hand, generative AI has reshuffled many things:
The consistency of the technological paradigm gives small teams a chance to compete directly against major companies.
The consistency of high costs forces everyone to think seriously about their business model, rather than simply relying on free offerings plus traffic to overwhelm the competition.
On the other hand, he is also very clear-eyed: the traffic and brand advantages of major companies still exert enormous pressure.
In the domestic market, small teams often need to survive first. They must either find an ecosystem player to attach themselves to, such as by becoming a key capability provider on a platform;
or find a sufficiently vertical niche and insert themselves into it through an exceptional user experience.
In overseas markets, by contrast, there are more opportunities to build momentum directly with paid products.
Because users are more accustomed to paying for clear value, even when it is a highly specific, niche tool.
Underlying this is a fundamental judgment:
As long as you genuinely solve a must-have need in a particular scenario, major companies may become your partners or customers, rather than enemies determined to swallow you up.
Today, BiYou Technology both provides technical capabilities to companies such as Kingsoft, Baidu, and 360, and operates ChatPPT and Utalk under its own brand.
This is in fact a very typical path for an AI vertical-application company:
developing products on one side while exporting capabilities on the other, growing its own niche ecosystem in the gaps.
VI. Pioneering Intelligence: It Is Time for Individuals to Take the Stage First
This year, many conferences have been discussing one keyword: the era of the individual. Under the theme “Pioneering Intelligence | The Era of the Individual,” Zhou Ze'an has his own interpretation.
He begins with the two characters kaichuang, or “pioneering.”
In his view, they represent a genuine breakthrough:
AI gives everyone the opportunity to start from their own scenario,
to build and operate a small but beautiful application, no longer being entirely constrained by platforms.
The era of the individual, meanwhile, signifies a change in how people participate:
We are no longer merely users of tools,
but can become designers of scenarios and creators of applications.
There are two particularly interesting formulations:
The individual is the scenario: every person and every small team represents a specific scenario. Once these scenarios are re-understood and re-encoded, they become a batch of highly niche yet enormously valuable applications.
The individual is the business: good businesses of the future will not necessarily all take the form of super-platforms. As long as one person or a small team achieves high ROI in an extremely narrow scenario, it is also a good business.
These are highly practical recommendations for implementation.
VII. Three Things Individuals and Small Teams Can Do in the Next 1–3 Years
If you are an individual or small team seeking to capture the AI dividend,
in his view, there are three things worth planning for in advance.
1. Shift your attention from technology back to scenarios
Rather than agonizing over whether I can code or whether I should train a model myself,
start instead with a question:
What specific scenario-based problem can I solve, and for whom?
Once the scenario is clear, tools and technology instead become interchangeable options.
The foundational models of the future will increasingly speak like people.
Technology will become easier to grasp, but gaining insight into use cases will become more demanding.
In other words, you do not have to be the person who understands models best, but you must be the person who understands the use case best.
2. Learn to dance with large models, rather than confront them head-on
Large models are like a main river that keeps growing.
What you need to do is connect tributaries at the right points, rather than try to re-dig a mother river.
This requires an awareness of boundaries:
Which problems will be naturally solved through the iteration of large models? Leave those to them; do not pour resources into the same direction.
Which problems lie in deep waters that large models struggle to reach? Highly specialized industry knowledge, highly specific business processes, and highly complex coordination within organizations, for example—these are precisely the areas where individuals have the greatest opportunity to invest further.
You should often ask yourself one question:
Am I addressing a large model's shortcomings, or competing with it for the same strengths?
3. Treat data as a long-term asset to be cultivated
Large models have reshaped the logic of computing power and algorithms,
but whether a problem can ultimately be solved in a given use case depends, to a great extent, on this:
Do you have that bucket of truly valuable data?
The data here is not a matter of more being better, but of greater depth being better:
the real Q&As, real documents, and real feedback accumulated through years of working in a field and serving users.
If you have been deeply involved in an industry for many years,
you may be holding the scarcest thing of all—use-case data and insights that others do not have and that are difficult to replicate.
Organize, clean, and structure that data,
then use open-source models and tools to fine-tune and test it,
and you will have a chance to build agents that are so genuinely useful in that use case that users cannot do without them.
AI Makes PPT Easier, and Makes Doing One Thing Well More Valuable
From 2015 to 2025, over the course of a decade,
many people changed tracks several times and took on several different identities;
Zhou Ze'an has always revolved around one protagonist that seems rather boring—documents.
Yet it is precisely this seemingly boring infrastructure
that determines whether we waste our days searching for features and aligning objects,
or spend them on genuinely valuable content creation and thought.
Whether ChatPPT or the personal super document workstation, they are essentially doing one thing:
pulling people out of the mire of tools, then gently nudging them back toward creation itself.
The AI era is not about making everything simpler;
it merely makes the things that should be simple simpler,
and then makes the one thing you truly persist in doing well more valuable.
Selected Interview Q&A
Q1: Please introduce yourself first. Who are you, and what does the company you are building do?
Zhou Ze'an: I am Zhou Ze'an, founder and CEO of BiYou Technology, and in a sense, someone who has repeatedly started businesses in the niche field of “AI + documents.” I have spent roughly thirteen years working around office documents: in 2015, I built an AI document product that was later acquired; afterward, I led the incubation and commercialization of WPS Smart Documents at Kingsoft; and in 2020, I left to start a business again, still focused on AI + documents. BiYou Technology is a team focused on AI document applications, currently refining products around “AI documents” and “document agents.”
Q2: What is the general background of BiYou Technology, and what stage has it reached?
Zhou Ze’an: BiYou Technology is essentially a vertical-application AI company, with its core team coming from Kingsoft’s smart-document incubation team. When we were founded in 2020, we first built a number of vertical document-model applications, such as AI résumé writing and AI contract writing, providing capabilities to platforms including Kingsoft, Tencent, and Shixiseng, and cumulatively serving more than 45 million users. In 2023, we shifted to a document-agent path combining “large models + small models,” launched ChatPPT, and have been fully committed to building a “personal super document office workstation” for the AI era. The company has also secured two rounds of financing, from Kingsoft Office and Baidu respectively.
Q3: What is your flagship product now? How would you summarize it in one sentence?
Zhou Ze'an: Our core product is ChatPPT, which can be understood as an “AI document-creation, end-to-end agent service” for individuals and enterprises. From content generation and structural design to layout and beautification, speaker notes, animation, roadshows, and then collaborative sharing, it seeks to cover the entire PPT-creation workflow.
Q4: What pain points does ChatPPT primarily help users solve?
Zhou Ze'an: Really, it comes down to four characters: save time, save mental effort. Traditional PPT tools have too many functions and too high a barrier to entry, so we simply retain only the “conversational creation” entry point, allowing users to state their needs in Chat and having the system invoke the functions automatically. That is the first step. The second step is that “generation must be trustworthy,” so we developed an in-house “traceable generation” capability that allows every slide's propositions to be traced back to the original document or webpage content. Third is layout beautification: we have our own Layouter model, which can rearrange layouts with one click and learn the user's own template style. The final piece is presentation delivery: we provide AI speaker notes, interactive animations, and AI roadshow capabilities, and we have also created integrated versions with Rokid and RayNeo that can project speaker notes into AI glasses for use as a teleprompter.
Q5: How do you view the developments in generative AI today that are genuinely breakthrough in significance?
Zhou Ze'an: I personally place greater weight on three points. First, open-source foundation models and distillation technology make it possible to deploy vertical use cases more lightly; it is no longer about whose model has more parameters, but about who is closer to the use case. Second, various Agent frameworks and protocols, such as MCP and A2A, give applications the ability to “talk” to one another; once they are connected, the pace of evolution will be much faster. Third, multimodality moves from text to images, voice, and video, upgrading the underlying modes of content expression as a whole. For a comprehensive medium like documents, this is an inevitable trend.
Q6: How have you genuinely integrated these frontier technologies into your products?
Zhou Ze'an: On the one hand, internally we use a typical “large models + small models” architecture. Of our 28 small models, roughly 60% have already been replaced and upgraded on the basis of open-source models, directly improving our service capabilities and delivery efficiency. On the other hand, in terms of Agent protocols, ChatPPT was among the relatively early products to embrace MCP; it has already been launched on major cloud platforms and in open-source communities, enabling other AI applications to access us in a lighter-weight way. In multimodality, functions such as “generate PPT from images,” “automatically add images to documents,” and “AI roadshow voice cloning” are all practical exercises.
Q7: In your view, how should technological innovation and commercial implementation be balanced?
Zhou Ze'an: I have three principles. First, “innovation must come first.” Especially in the AI era, without innovation, you are basically not qualified to compete. But this innovation does not necessarily have to be in underlying algorithms; it can also be innovation in application technologies within a use case. Second, “business must be validated.” No matter how cool something looks, if it cannot run through a commercial closed loop in real-world use cases, its value to a company is limited. Third, the two should not be separated; instead, their weighting should be dynamically adjusted according to the stage: in the early stage, emphasize technological innovation plus a small amount of commercial validation; later on, strengthen commercial validation, which in turn drives technological optimization.
Q8: From a global perspective, what are the opportunities and challenges for AI office products?
Zhou Ze'an: There are two opportunities. The technology paradigm is consistent: everyone is using similar large-model systems, so small teams have an opportunity to directly benchmark traditional giants in niche use cases. The cost structure is also more consistent: AI applications inherently entail costs such as computing power, which instead forces everyone to seriously consider the “product + business” combination, making a completely one-sided squeeze through free offerings less likely to recur. There are mainly two challenges: first, the traffic and brand impact of major companies remains very strong, especially in China; if they rely only on ToC traffic, startup teams face tremendous pressure. Second, technology is highly dynamic. If early technology and product choices are too superficial, overlapping with or being replaced by the direction in which large models are developing, they will be left in a very passive position later. Strategically, in China the priority is more often to find a niche and survive—for example, by “parasitizing” the ecosystem of certain platforms—before gradually becoming independent and scaling up; overseas, there are more opportunities to launch directly with a clear paid model.