At a time when the AI boom is sweeping the world, the domestic AI industry has fallen into a vicious circle of “seemingly prosperous, but actually anxious”. There are models and computing power, but the application has always been difficult to implement, and the user experience is frequently “overturned”. What’s the problem? Is the technology not strong enough, or is it going in the wrong direction? This article will take you to peel back the façade and face the “real questions” that have been overlooked but are crucial – about products, about ecology, and about how we can really use AI well.
The biggest difference between overseas and domestic AI may not be the level of the model, but really not applied. This leads to a terrible consequence: the entire domestic AI ecosystem is broken.
What is a broken chain?
The image is that the computing power is the computing power, the model is the model, and then the end user does not close the loop.
For example, e-commerce in the past was ecological:
Some people do clouds, some people do e-commerce platforms, some people do suppliers, and there are consumers, so that the technical integrity of the Internet and the business system are integrated, and everyone in the chain can have benefits.
Now that AI is almost running in the overseas positive feedback chain, at least on the B-side, which I perceive more clearly, things are relatively clear:
Nvidia provides GPUs, model companies provide infrastructure, application companies provide SaaS integrated with AI, and B-end companies gain efficiency.
Once the ecology forms positive feedback, it will allow all parts to link and tend to develop benignly, breaking the curse of AI development in the past 10 years.
If you don’t close the loop, you have to rely on investment, how can there be so much investment now…
If you want to say what will most likely lead to the overall backwardness of AI in our country, I will most likely choose this.
Why is it more difficult to apply AI in China?
The core reason is that we have to skip a relatively easy stage and integrate the business more directly from end to end (“Unmanned Company” is actually writing about this new model).
AI is now a better tool in many scenarios, and the level of intelligence is roughly sufficient, but the challenge behind integrating complete services is great.
To answer why one stage must be skipped, we have to talk about SaaS.
SaaS can be seen as the area where AI is most likely to exert its effectiveness and generate immediate business value.
The problem is that there is no SaaS in China (it cannot be said to be 0, but there is no SaaS market of the same size overseas).
At this time today, I think that anyone who has done tools and SaaS needs to face a particularly cruel inner torture:
Will there really be SaaS in the future in China? Is AI good or bad for classic SaaS?
If you are rational enough, even though it is very painful, you still have to answer:
No matter how many tubes you put on it, how many flowers you put on it, it is just a decoration for the tombstone. From my point of view, the so-called delivery result means a few more tubes, but it cannot get rid of the objective law of medicine and medicine.
The market environment of ten years of verification will not change, and the technical conditions are making you lose bargaining power (large models will make it easier for SaaS Party A to develop themselves), how can such a model not be hung up!
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
The key is that the integration of SaaS is actually low, so it is easy.
In the context of a company, SaaS always solves a small problem, so it is destined to deal with low complexity. The tools for handling recruitment must be less complex than completing the entire company’s business.
Now this part of the low complexity has no commercial value, so we must face the complexity of the business, so we must skip a stage.
End-to-end business integration
The book “Unmanned Company” is not about people and no people, but about the challenges brought by this high-complexity scene that we must face.
Recently, in my speech on “Unmanned Company”, I have been mentioning a key issue: our value creators have been migrating. And this migration of AI may be much more thorough than in the past.
This can be found in various industries, this time we will flash back, let’s use the recent Antropic ProjectVend as an example. Understanding this migration means understanding end-to-end business integration.
The Vend project is not complicated to say, it is to completely use AI to operate the following container, including the goods, how much to sell, etc.:
Although unfortunately, after giving it $1,000 in start-up capital, it went bankrupt due to poor management. But its pattern is very different from the past:
Comparing it with the e-commerce platform, the obvious degree of automation has been improved, it is AI that is running this small store, rather than simply building a platform, and I will help you find out who buys things.
So compared with the past e-commerce platform or online store on the e-commerce platform, it is not a thing in nature.
The e-commerce platform is actually more advanced than the past department stores, which relies on algorithms for a large part of distribution and transactions, including what you are looking for, the status of logistics, etc.
So if you put the Vend on the 2000-year-old department store – e-commerce platform and online store – on an evolutionary route, what will you find?
The theme of value creation is increasingly shifting towards silicon-based.
Behind this is end-to-end business integration based on AI (not necessarily an end-to-end model).
This is the biggest background of the times we have to face, and it is also what “Unmanned Company” wants to say.
Too many books talk about what a large model is, how to write prompts, etc. In fact, we should pay more attention to what is the AI model that can run through in China, otherwise we are more likely to miss AI when embracing AI, just like providing web services when the Internet is provided.
Suppressing increased complexity
Because SaaS is completely cold, we don’t have the opportunity to deal with low-complexity scenarios, and we have to climb steep slopes to deal with high-complexity scenarios. (The legendary northern slope of the Himalayas)
Of course, at this time, you can still choose between improving or building a new one.
It is difficult for existing companies to completely throw away the existing mess and start over. What is the essential meaning at this time?
Everyone who wants to improve has to build a product that is more complex than Glean, a company now valued at $7 billion.
If you look closely, you will find that in such a system, business knowledge and AI are seamlessly integrated. In other words, it is necessary to have a complete understanding of the business and AI to complete this improvement.
Digitalization has been around for more than 10 years, and now it has increased the difficulty brought by AI, which is obviously more challenging.
Recently, the Harvard Business Review published an article titled “Li Ning Digital Intelligence: What is the Experience of Having a Morning Party with a Digital Store Manager?” (the picture below is from this article), from which we can clearly feel that the complexity of this system is far greater than that of SaaS.
You can think about whether such a system is complex or a single point SaaS is complex.
And this complexity is often endogenous in the business, if it is really a hard resistance, relying on the progress of technology (including AI) to throughput this complexity, the typical problem at this time is obvious: AI is actually not easy to use, and the current level cannot solve the business of this complexity.
How can complexity be reduced at this time?
If we think that complexity comes from two sources, one is endogenous to the business and the other is caused by backward compatibility, then the core way to reduce complexity can only be non-backward compatibility.
This is not true in all areas, but if you can find it, you are obviously lucky.
But even if you find it, it will definitely be more complicated than just making a tool.
So back to our title: the difficulty of doing AI applications in China is definitely underestimated.
Skill set
What skill set is suitable for the above products? Interestingly, Glean’s resume can be found.
This guy has been doing architecture for a long time, and the person who really does architecture is often a big comprehensive, and he must have an understanding of business, technology, etc., and then he can organically pinch them together. Of course, I have to add a point now: I have to understand AI, after all, the technical characteristics of AI are completely different from the traditional software technology stack.
In abstract terms, it means that the complexity of the fields covered by your product will increase, which in turn will require traders to improve their cognition and control ability of complex systems. Pure scientists or CEOs who don’t understand technology and are good at business may not be able to operate products like Glean.
The most critical role in a human-centered system must be to manage people, and now the new system needs people who better understand people, AI and business because of the migration of value creation subjects. If it is an extreme value, that is, an unmanned company, then it is probably enough to understand AI and business.
This is another new challenge.
brief summary
It is obviously more difficult to do real AI applications in China, but the skills that rely on it are relatively simple to do are similar to those in the past, but relying on it obviously cannot solve the problem of broken chains.
Without a real application ecosystem, the development of AI is bound to be unsustainable. Since 2010, we have proven this for more than ten years, so let’s not prove it again.
To exaggerate further, the AI competition between China and the United States must be a long-term competition, and the long-term competition may not depend on one city and one place, but on whether the ecology is benign enough.
So this is a real problem.
