The ultimate question in AI entrepreneurship: Will tech giants understand innovation first, or will startups gain access to channels first?

God Translation Bureau is a translation team under 36Kr, focusing on fields such as technology, business, the workplace, and life, and mainly introducing new technologies, new ideas, and new trends from abroad.

Editor's note: The hype around AI is deafening, but the home screen of your phone reveals the truth. The old rules of entrepreneurship may not be outdated yet. Will established giants acquire the ability to innovate first, or will startups gain access to distribution channels first? This article is from a compilation.

There is still much we don't know. Here are some relevant questions.

The Home Screen Test

Although this is the golden age of artificial intelligence, we are still in the very early stages. To illustrate my point, let's conduct a "home screen test":

How many apps on your phone's home screen are "AI-native"? And among them, how many were created using AI programming tools?

Strangely enough, for most of us, this number may be closer to zero than you think, especially after excluding those obvious large language model apps (such as ChatGPT, Grok, etc.). Of course, many people are working hard on this, but as of now, the result of the "home screen test" is that there are almost no AI-native apps in the 4x7 grid on our screens. Shouldn't the future be that all 28 apps are AI-native? Where is the AI-native calendar app? And where is the AI-native social network? I remember how quickly my habits of using instant messaging, social, and email apps shifted to the mobile app versions during the wave of the transition from the web to the mobile. But now, what exactly is going on?

This tells us that there are huge opportunities ahead because we haven't really started using AI to change the way we work. So far, the only personal change I've made is to reduce the number of Google searches and increase the frequency of conversations with AI. But obviously, the potential of AI goes far beyond that. I've explored "ambient programming" and the network effects in the AI world in more detail in other articles.

Some Other Important Questions about AI and How It Will Change Entrepreneurship

I believe that not only are AI products themselves in the early stages, but our thinking about "how to build these products" is also at a very elementary stage. That is to say, AI will change the way we create companies, but it's not clear yet how this change will occur.

Here are some outstanding questions, and people can have very different but reasonable views on them:

• Will future startups need fewer (or more) employees? One view is that AI may bring a 1000-fold leverage effect. Therefore, the work that used to require an entire company to complete can now be done by one person. Naturally, this may mean that one person can supervise 1000 AI agents that code around the clock and build a billion-dollar company in the process. That's certainly amazing, but the counter-argument is: If an AI-native startup can expand rapidly, but some of its business capabilities still need to be done manually because AI can't handle them yet, then it's foreseeable that they will still need to hire a large number of employees. Maybe the bottleneck will ultimately lie in "taste" (as some people say), causing the company to need a large number of designers. Or perhaps you can make prototypes of many popular products very well, but...

• How will companies build moats in the face of fast-moving AI competitors? If a product can be easily replicated instantly, and AI homogenizes all functions, how do you define your barriers? A simple answer is that the experience of consumer apps in the past decade has taught us that in an ecosystem with little technological differentiation, user growth and network effects are everything. On the other hand, the development speed of AI software may be so fast that the ability to "continuously iterate, release new features, and create new products" itself becomes the key to differentiation. One of my thoughts is that the only products worth developing will be those that require years of planning and large capital expenditures. For example, space technology or B2B hardware - these fields require you to have a deep intuition about the current market to place the initial bet. The advantages of any software product that AI can develop within a few years will be offset by arbitrage. The advantages in these fields will be more like those in the direct-to-consumer (DTC) field, where brand and insight into channel timing are crucial. But it's much more difficult to build a great company in this way.

• Will AI make entrepreneurship cheaper or more expensive? Obviously, infrastructure layers like foundational models require a huge amount of capital investment. In theory, the development of certain applications should become very easy. However, growth and distribution require a large amount of money to get the product into the hands of customers. One thing we've learned in the past decade is that even though the cost of developing a web application is relatively low, the cost of acquiring users can still be as high as millions of dollars. The high failure rate is because there are too many products competing for people's attention, so this may be the limiting factor.

• How will we organize future teams? Does it still make sense to keep independent functions such as engineering, product, and design? Or, when AI technology can develop software based on product requirement documents (PRD), wireframes, or any form of input, making product development completely multimodal, will all these disciplines merge? Many amateur history enthusiasts know that the organizational structure of work has changed greatly in recent centuries. We've shifted from the model of family workshops (where family members produce goods by hand) to the model of factories and companies as industrialization progressed. Recently, software platforms have given rise to a large number of independent contractors and part-time workers, such as food delivery riders. It's not clear yet how AI will affect the next generation of work models.

• Is the position of the San Francisco Bay Area as a technology center still solid? In the past, Silicon Valley became the center of entrepreneurship because of the network effects it had in terms of talent, venture capital, and knowledge. I think that although the advantages of San Francisco are still strong, as a large number of talents flow to other centers such as New York, and knowledge is more widely shared through channels such as podcasts and Substack, this advantage has undoubtedly weakened. The argument in support of continuous decentralization is that if developing a product becomes extremely simple, it almost becomes a form of content creation. As we've seen with content creators, they can be local or global and can be based anywhere. If you no longer need to hire employees or attract venture capital to scale up, then future entrepreneurs may also be spread all over the world.

• How will venture capital firms operate in a world where startups are fragmented? For a long time, the model of venture capital was simple: invest in excellent startups that emerged from Stanford University and the surrounding towns where tech talents gathered. What will happen when it becomes extremely easy to develop a brand-new product and test it in the market? If people are willing to pay for these products, they may be profitable from day one, especially if they are developed by only one or two people at a very low cost. What will happen if such products are continuously created all over the world? Perhaps venture capital itself will become more decentralized because it will be used as growth capital rather than risk capital. This will make it clearer and easier for a wider range of investors to participate, even if these investors are spread all over the world, not just limited to the Bay Area.

• Related to this, will the divisions of pre-seed, seed, and Series A/B/C financing rounds still exist? We use these distinctions to divide different growth stages. But maybe in the future, more products will go directly from zero to Series A. So, does it still make sense to fund people to conduct various explorations and experiments? Or will all of this happen in the form of side projects?

The beauty of these questions is that the history of the contemporary technology industry is only a few decades, not centuries. Many modern structures, such as venture capital, the mobile Internet, or starting a company in downtown San Francisco (rather than on the Peninsula), are relatively recent phenomena. Because everything is changing so rapidly, it's easy for us to imagine that it will change again in the next few years.

Historical Precedents

Once you start thinking about how business structures evolve step by step with technological development, the above questions seem to be obvious directions for exploration. Recall the cottage industry in pre-industrial Britain. In the 18th century, a blacksmith might work in a workshop connected to his home, with the whole family involved in production, and the final products were made one by one and sold in small batches in the market. Fast forward to the 19th century, industrialization meant that a large number of laborers worked in factories. At that time, you needed a company to organize business activities, shareholders to provide funds for expenses, and layers of specialized management. (You can read Burnham's "Managerial Revolution", which has a very interesting discussion on these dynamics.)

Or you can also look at the 17th century. With the development of ocean - going naval forces in the form of merchant fleets and intercontinental trade networks, limited liability companies emerged to organize all of this. The East India Company established the largest company in the world at that time and had a standing army of 260,000 people through both technological and business model innovations.

Therefore, when you examine the potential impact of AI, it seems inevitable that the business structures we've created today will ultimately be insufficient to organize all the potential productivity brought about by this technology. If we've evolved from "organizing labor by family units" to "organizing labor by factories", what does this mean in a world composed of AI agents, computing power, and models?

The Most Optimistic View

In the most optimistic scenario, I think we hope to see a world where AI enables fewer people to create more value. Therefore, it would be great if an AI-native startup could be founded with only a very small number of employees. A company's moat will be built through killer features and technology, rather than simply relying on distribution channels and monopoly advantages. (We want more new things, rather than having established giants dominate the next generation.) We hope that all these new technologies will make the cost of entrepreneurship lower. Ideally, the Bay Area will continue to be the technology center. Even if people can establish companies anywhere, they will still move to San Francisco to take advantage of the local expertise and capital. Of course, venture capital will also evolve and find ways to profit from it :) I hope this is true.

But for each of the above points, you can also put forward a valid counter - argument. The biggest winners in AI may be those companies that have large data centers, massive amounts of data, and powerful computing power. This means centralization, and large companies will become even larger, creating a world that is unfavorable to startups. Or, AI may just be a great set of features, but it doesn't help much with marketing and distribution. Existing giants can slowly transform their original products into products with AI-native features and ultimately defeat startups in the competition.

As some smart people I know have said, the question may be: "Will established giants acquire the ability to innovate first, or will startups gain access to distribution channels first?" The established giants may win.

In the next few years, the trends in the startup circle will lay the foundation for the future development pattern. The past few years have been eventful. We've seen a number of top - notch AI research teams developing foundational models. But now, these models have absorbed all the data in the world, and their effects are gradually reaching the limit. The protagonists in the next few years will be those who build business logic on top of these models. They won't conduct AI research or train their own foundational models. Instead, they will adopt a model - independent strategy and create an engaging user interface at the top level. We're seeing such products being integrated into various vertical industries, from sales tools to integrating entire industries, in a variety of ways, and applying technology in this way.

The next few years may be full of changes.

Translator: boxi.