Wang Daoping of China Creation Ventures: Many AI products are abandoned by users as soon as they are launched. It's very cruel.

This is a joint column launched by Pencil News and Global AI, exploring the mainstream opportunities in AI applications. Global AI is seeking to invest in early-stage AI projects. Submit your business plan at: https://globalai.hanghang.com/

Interview | Zou Wei, Zhao Nan

Last year, Sam Altman, the founder of OpenAI, put forward a startling view: With the help of AI, a founder can build a unicorn company (valued at $1 billion) without hiring any employees.

In the eyes of many in the technology industry, this has already become a trend, and it's just a matter of time.

Since ChatGPT was launched in November 2022, the popular startup directions in the AI track have changed several times. From the high-threshold basic large language models, to various AI "shell" applications trying to transform traditional industries, to multi-modal and intelligent agents, and recently to AI-native products and new interaction paradigms.

It is generally believed that AI-native products and new interaction paradigms are the directions most likely to give birth to "one-person unicorns". They are no longer adding AI to old software, but building AI-centered products from scratch. There have emerged AI personal multi-modal assistants (such as the Fellou browser) and AI-native hardware (such as Even glasses and Plaud recording pens).

Recently, Pencil News had an exchange with Wang Daoping, the founding partner of Huachuang Capital, on topics such as the latest AI startup opportunities, the capabilities of AI entrepreneurs, and AI commercialization.

Wang Daoping believes that the time left for AI startups is very short. Once users see the results, they will immediately form a judgment on your product, and the tolerance is extremely low. Moreover, as soon as the prototype of a good product appears, large companies will quickly notice and follow up. The pressure this brings to entrepreneurs is unprecedented. Therefore, AI product entrepreneurs should figure out from the beginning what core problems the product solves, and even consider the commercialization path.

Contrary to the traditional impression, when AI becomes a "consensus track", financing becomes more difficult, and the requirements for AI entrepreneurs are even higher. They must quickly create differentiation and achieve a large enough scale to avoid being replaced.

He particularly emphasized that AI entrepreneurs should not shy away from talking about commercialization. "Commercialization doesn't mean you have to make money on the first day, but you at least have to prove that the product 'works' and that someone is using it. This is very crucial."

Recently, Wang Daoping also joined the investor lineup of Global AI, which is an ultra-early investment acceleration platform for global AI application entrepreneurs, aiming to promote the arrival of the era of "one-person AI unicorns".

The following are the highlights of the interview.

Statement: The interviewee has confirmed that the information in the article is true and accurate. Pencil News is willing to endorse the content.

The underlying AI technology is still evolving. Optimistic about general-purpose applications

Pencil News: Since ChatGPT was launched in November 2022, what major changes have taken place in the startup opportunities in the AI track?

Wang Daoping: It changes almost every year. Mainly because the AI technology itself is still in a stage of rapid development and has not entered a "convergence period" or "stabilization period". Whether it's the technology or the application form, many things are still uncertain.

We started paying attention to this track around the end of 2022. When ChatGPT came out, the first thing we looked at was the wave of "large models", and our focus was still on the models themselves, including the underlying infrastructure.

In 2023, we invested in Dify. It can be understood as an intermediate layer above the model, facing developers, helping them better use large models or related tools.

During that period, we also paid more attention to Infra (infrastructure)-related opportunities within the company, but not the most basic ones, but some ability expansions based on large models. As for the earliest batch of large model projects, although we looked at some, we didn't invest in the end. On the one hand, the valuations were indeed relatively high at that time; on the other hand, we thought the model technology was not mature enough, and the commercialization path was not clear enough.

Moreover, the field was already very hot at that time, and large companies were all involved. Whether it was OpenAI or several domestic giants, they had basically all laid out in the model layer. So we could also feel that the space for startups in this direction was actually limited, and the competitive pressure was great.

Later, we also invested in a project called "Silicon-based Mobility" from the end of last year to the beginning of this year. This is our continuous follow-up on the Infra layer opportunities.

This year, new trends have emerged, such as "intelligent agents". The number of To B and To C applications related to intelligent agents has increased significantly, and there are even many general-purpose directions.

For example, the emergence of Manus is very representative and a landmark event. Now we can see that more and more entrepreneurs are moving towards the intelligent agent direction, which also shows that this direction is evolving rapidly.

In addition, there is another direction called "AI + hardware". We also invested in AI glasses (Even) last year. This year, we have also seen many new opportunities in the directions of robots and terminal devices.

So overall, the startup opportunities in the AI track are evolving dynamically, with different focuses each year. The underlying technology has not converged yet, and new directions may emerge next year. There are opportunities emerging at both the software and hardware ends.

Pencil News: Which industries do you think AI will combine well with? (Medical, education, finance?)

Wang Daoping: Personally, I'm more optimistic about relatively general-purpose and highly market-oriented industries. For example, the consumer field, and some To B general-purpose tool directions. Of course, there are also opportunities in To C. The medical and education directions you mentioned, I think there are indeed opportunities, but these industries are quite special, and there are many problems that AI can't solve immediately.

We have indeed seen many AI startup projects in industries like medical, education, and finance, but the challenges will be more and more complex. It's not just a matter of product and technology, but also factors such as industry rules, compliance, customer acquisition, and data privacy. These all bring considerable pressure to entrepreneurs.

In these fields, you may only realize during the startup process that "oh, there are actually so many hidden thresholds here", and these can't be solved by just product or technology. So we tend to focus more on directions with relatively open tracks, clear commercial paths, and rapid user feedback.

Pencil News: What aspects do you value most when investing in AI projects?

Wang Daoping: We mainly invest in early-stage projects, so the most core thing is to look at the team. One is the team's insight into AI, that is, their intuition and judgment in understanding, using, and creating AI products, which includes productization and commercialization capabilities. Since many of the projects we invest in are on the application side, product capabilities are particularly crucial.

For example, in the AI glasses project we invested in, the founder himself is a very excellent product manager. He not only has strong product capabilities but also has a very accurate judgment of this track.

Although these are startups, their observation and understanding of the track are actually at a relatively advanced level in the industry. So we attach great importance to the team's grasp of the opportunities in the track they're in and their ability to create something with product significance.

Pencil News: Are there any AI startup directions that seem very popular now but you won't invest in?

Wang Daoping: Large models. The capital consumption far exceeds what an ordinary startup can bear.

From the very beginning, its business model is very unclear. This direction is often technology-driven, and both the product form and the monetization path are very vague.

Moreover, the large model track itself is highly consensus-driven, and everyone knows it's an important direction. In this case, large companies naturally have more advantages. Whether it's talent reserves or resource integration, they can quickly invest and advance. Startups have almost no chance to "surpass" them in this regard.

Even if you, as a startup, create a preliminary product, this path won't "converge quickly". You have to keep investing and financing continuously. But if your commercialization ability can't keep up, you may not be able to raise the next round of funds, and the company won't be able to go on. This is a very real problem. For startups, financing is often stage-by-stage, not raising enough money for three to five years at once.

Also, in such a very popular track, it's hard to negotiate a reasonable valuation. For our investment institutions, we also have to consider ROI (return on investment). If the commercialization path is unclear and the investment is huge, it will also bring great pressure to us.

Wang Daoping, founding partner of Huachuang Capital

AI entrepreneurs should create revenue as early as possible

Pencil News: For today's AI entrepreneurs, are there any differences in product capabilities compared to what was required in the mobile Internet era?

Wang Daoping: There are several obvious changes.

First of all, the speed requirement for AI startups is now extremely high. Unlike in the mobile Internet era, where you might have one or two rounds of iteration time, today many AI products need to show results as soon as they're launched. Because AI is essentially an "instant" productivity tool, once users see the results, they will immediately form a judgment on your product, and the tolerance is very low.

This results in less room for product trial and error for entrepreneurs. Once it fails to meet expectations, it will quickly be abandoned by users. This is the biggest difference from the past.

Another significant difference is that the competition rhythm has accelerated. Now if you create a promising product, even if it's just a prototype, large companies will quickly notice and follow up. Even if the products they create are just "similar", they can quickly expand with their resources, channels, and user base. This brings great pressure to startups.

In the mobile Internet era, you might have a so-called "quiet period" for small-scale testing and continuous iteration. But in the AI field, there is basically no such window period now.

Today's AI products emphasize direct result orientation. Unlike before, where you could first focus on user acquisition, retention, and then consider monetization. Now you have to solve problems from the start and even consider the commercialization path. As soon as the product is launched, the capabilities of the entire team, such as commercialization ability, resource integration ability, and financing ability, must be immediately demonstrated.

So today's requirements for AI entrepreneurs are not only about strong product capabilities but also a comprehensive test of team building, execution efficiency, and resource coordination ability.

Of course, some basic skills of the product itself, such as user insight and function refinement, still have something in common with the past. But the overall environment and rhythm are indeed very different from the mobile Internet era.

Pencil News: You mentioned earlier that a key point for AI projects is to have an impact on users as soon as possible, and the other is to achieve commercialization as soon as possible. But many AI projects in the industry still have difficulty in commercialization. What do you think of this problem?

Wang Daoping: It's very challenging. Although commercialization doesn't mean you have to make money on the first day, you at least have to prove that the product "works" and that someone is using it. This is very crucial.

I think we're still in an early stage, and the development of the entire industry is not yet finalized, and no clear pattern has been formed. Just like when the mobile Internet emerged, you could see some iconic good products, such as excellent social applications, communities, and e-commerce platforms. But in the AI field today, we haven't seen clear "winners" or mature models yet, and many are still in the exploration stage.

However, one thing is certain: a good AI product must have a smooth closed-loop operation, that is, the product can be continuously used by users, and they are even willing to pay for it. Now many intelligent agent (agent) products or other AI companies choose to go overseas to target the overseas market because overseas users have more mature payment habits, and the commercial closed-loop is easier to run smoothly.

In contrast, there are obviously many problems in the domestic market. For example, large company products like "Doubao" are free, and even free in the long term. As a startup, if you also target the domestic market, you immediately face a real problem: how do you compete with Doubao?

You can choose to be free too, but then how does the company's future business model work? How do you clarify your commercialization logic? What are your expectations? These are questions you must explain clearly.

Of course, it doesn't mean that targeting the domestic market won't work. You just need to come up with a convincing path to prove how you can achieve sustainable development in such an environment.

Pencil News: In the mobile Internet era, many successful startups actually raised funds in multiple rounds. Will this situation happen again in AI startups? After all, the capital in the market doesn't seem as abundant as it was before.

Wang Daoping: It's indeed possible, but it can't be generalized. In fact, when we look at many large companies, they didn't raise a lot of money in the early stage. For example, ByteDance and Pinduoduo didn't need a huge amount of financing from the start of the business until they could achieve self-sufficiency. Of course, if they expand later and then raise funds, that's a different stage.

I think one thing is very crucial, which is that the capital supply in the market has indeed decreased today. Whether it's US dollar funds or RMB funds, there isn't much market-oriented capital. In China, a lot of the current capital actually has a government background, such as state-owned assets, central enterprises, or some industrial funds. Their investment logic when investing in projects is different from that of market-oriented funds. They pay more attention to industry orientation and strategic goals rather than pure market opportunities.

As VCs, we used to invest wherever there were market opportunities, but now many funds have obvious policy orientations or industry preferences. This creates a different financing environment for startups.

Moreover, AI is now a "consensus track", which is different from a "non-consensus track". If you're in a track that everyone recognizes, theoretically, financing should be smoother. But the problem is that it's actually very difficult to create differentiation and achieve a large enough scale in this track. At this time, the challenge may not be financing but whether you can truly achieve scale and form a moat.

Pencil News: For today's AI application entrepreneurs, does it mean they need to learn to create revenue earlier? Because there is less capital in the market, and it's difficult to rely on multiple rounds of financing to "survive" until the business model matures.

Wang Daoping: That's indeed the case. Today's entrepreneurs need to solve the revenue problem earlier.

However, the good thing now is that the team size is generally getting smaller, so the startup cost is relatively more controllable.

On the other hand, the current financing market is also different from the past. The sources and nature of capital have become more diverse. Some companies will seek industrial investment, such as investment from large companies. This is also a way. In addition to capital, this way may also bring support in channels, resources, and capabilities. Whether entrepreneurs can find the right type of capital is also a question they need to consider.

Of course, it would be even more ideal if you can achieve commercialization faster. For example, Plaud, which makes AI recording pens, makes money directly from the product without relying on financing. This is a good approach. This path is worth referring to.

But not all companies can do this. Some companies, such as those focusing on robots or commercial hardware, may not be able to make a profit in the short term. Or the market maturity is not enough, so they have to find ways to solve the financing and survival problems during the development process.

The underlying logic of AI intelligent agents

Pencil News: Now many people are paying attention to startups in the intelligent agent (agent) field. In which directions do such entrepreneurs have more opportunities?

Wang Daoping: I think there are opportunities