In the Internet age, we are familiar with a term called PMF (Product Market Fit), which refers to the best fit between the product and the market. The product you provide just meets market demand and satisfies customers. This is the first step to entrepreneurial success.

With the rise of AI big model technology, we began to hear a new term - TMF (Technology Market Fit), which emphasizes that we need to match technology with the final market. This concept highlights the importance of understanding and cognition of technology in the current technological transformation.

After several years of rapid development of AI big model technology, let us explore in depth: Why did PMF evolve to TMF? What is the essential difference between the two? Does TMF also have potential problems?

Why Technology Matters

• Fuzziness

A notable feature of this round of AI big model technology is its high degree of ambiguity. This ambiguity is reflected in: the same input may get different outputs; a demand that cannot be solved today may be solved directly in three months. This ambiguity based on probabilistic prediction and rapid development has brought us huge challenges in product construction. Because as a product provider, we need to provide end users with certainty, and there is a natural gap in between that needs to be bridged.

Imagine, as a user, if you perform exactly the same operation when using the product, it works today but not tomorrow, which is obviously unacceptable. From the perspective of product technical solution design, you may have spent a lot of effort to solve the current problem in an engineering way, but found that the model update automatically solved it three months later. This requires us to think more deeply about the value and significance of the current solution. Therefore, the ambiguity and unpredictability of this output and technological development require AI product developers to pay more attention to the technology itself. You need to understand the boundaries of technology and be able to basically judge the development trend of technology, such as which problems may be solved and which problems are more difficult to overcome in the short term.

• Transformative

The second important feature is the revolutionary nature of the technology. This mainly refers to the fact that AI large model-related technologies can solve some problems that we were completely unable to solve before, or that are extremely difficult and costly to solve. Imagine how difficult it would be to apply a style transfer to an image before ChatGPT-4.0; and even further back, before the emergence of Stable Diffusion, how arduous this task would be.

This technological revolution may have a dimensionality reduction impact on many existing products and services. The business and economic models we are accustomed to may change, and even bring about revolutionary social changes. Therefore, we have to pay attention to the importance of technology and think about: With the support of new technologies, are many of the things we do still meaningful? What is the new value? What other value-added services can we provide?

• Versatility

The last point is versatility. AI big model technology can understand natural language, and now many agents can also operate computers, mobile phones, browsers, etc. Imagine if an assistant who is much smarter than you can use a variety of tools, which is really shocking.

This versatility means that in theory AI can do everything you can do, or even more. In some areas where you are not so professional, AI may perform better than you. This prompts us to rethink the value of the product to users and the role of technology in it.

The user is always the starting point

Due to these technical characteristics, some people have proposed the so-called TMF concept. However, as more and more AI products are released and people's understanding of AI continues to deepen, we find that: although AI technology is important, for actual products, users should always be the initial starting point.

The recognition and understanding of technology should be the resource and path for us to make products, rather than the purpose itself. No matter how technology develops, the core is always to solve the problems of specific users in specific scenarios.

Start With Problems, Not Technology.The goal isn't to showcase AI. It's to solve problems so seamlessly users barely notice the technology at work.

"Putting users first" seems to be something everyone can say, but it is difficult to really do it. Each of us has our own ego, just like saying "I understand you" when comforting others, but in fact we know that we can never fully empathize with them.

Back to the user, we need to think about a series of questions:

• Who are your users? Who do you want to provide services to? How do you plan to deal with those who are not your target users?
• What is the user's current scenario? When and where does the product appear?
• What is the specific problem? What is the user's goal?
• What do users see? What do they feel? What results do they expect to see after taking each action?
• What is the end goal of the user? What is the solution you provide? What problems can and cannot be solved?

Users are not static. Before and after using your product, users' cognition and understanding will be different; users who have been exposed to competing products and those who have not been exposed to competing products also have different expectations of your product. Changes in the external environment and information continue to affect users' cognition and understanding.

Therefore, demand is infinite. Users are a collection of demands. Changes in users will always lead to unmet demands, while currently met demands may trigger new demands and changes.

User experience is more important

In the AI ​​product environment, user experience becomes more important. The page is simpler, and we no longer need so many buttons and options to interact with users. Users can directly express their needs in natural language. However, there are also traps here: although natural language interaction has the lowest threshold, is it the most efficient? Can users express their needs accurately and clearly in natural language? Is the display of text clearer and more efficient? These issues require us to think deeply. The page is simplified, but the requirements for details are higher. Details are the devil. When the content presented on the page is reduced, the details and slight differences in user experience will be magnified. The entire simplification is not just a reduction in page elements, the interactive experience may also be simpler. Users may get the final result directly after making a request, so how can the intermediate process better meet user expectations? How should the waiting process be designed? Should users wait for the result, or can they perform other operations at the same time?

The essence of user experience design is to design technology based on user needs . Some AI product developers will blame users for not being able to use their products, thinking that users do not understand the technical features. But we should ask ourselves: Why do users need to understand the technical features and boundaries?

The essence of product and experience design is to solve problems that may exist in the process of technology meeting user needs, rather than problems with the technology itself.

When designing AI products, the goal is not to demonstrate all the features of the technology (unless you are selling the big model itself), but to leverage the strengths of the technology based on user expectations and needs, and to meet or even exceed user expectations.

In the current situation where the performance of different large model technologies is not much different, user experience naturally becomes a difference point that users can more easily perceive. Most products are calling APIs and using the latest models, whether open source or closed source. Therefore, the difference in product experience mainly comes from the design method, including interaction design, the way of exposing model capabilities, system prompt word design, and how to combine user input with relevant background and context information to obtain more expected responses. Using similar technology to create different experiences is the key. Of course, there are special cases. For example, when DeepSeek R1 was released, most people may not have used inference models, and inference models can indeed give users a completely different experience from the model level. Therefore, pursuing higher intelligence is also a path, especially when the level of intelligence provided by large models is significantly higher than all current models, and users can perceive this significant improvement.

Finally, due to the versatility of technology, we may need to do some subtraction in specific user scenarios, that is, to impose some restrictions on the capabilities of large models or set preferences. Technology is vague, but scenarios and problems are clearer, and we need products to provide users with more certainty .

The core essence of experience is to provide certainty, including certainty in the process and certainty in the results. Users may expect the product to provide an experience beyond their expectations, but this experience beyond expectations also needs to be based on certainty. After clicking on the product, the expected effect appears, and the output basically meets the expected results. We need to have certainty first before we can have subsequent surprises beyond expectations.

Of course, some products may be positioned to provide random effects, such as conversation companionship or entertainment applications. But if you think about it carefully, these scenarios also have constraints and boundaries. It is not just randomness that can bring a good experience. It also requires careful design and thinking.

To sum up

We can use war to analogize the relationship between technology and AI products: all strategies should be formulated based on the opponent, and weapons should be selected according to the opponent, rather than stubbornly using your own weapons - that is showing off, not war. All technologies are tools and resources, not the purpose; the purpose is your users, their scenarios and problems, and the value you provide.

TMF is valued because technology is a very important variable, but what remains unchanged are the market, the users, the problems you hope to solve and the value you provide to users.

You need to find a way to meet the expectations and needs of users. Technology is not the user's expectations, but the products and services you provide through technology are.

Finally, this is indeed a difficult thing. This is probably something everyone understands. If you ask anyone who is starting an AI business, they will probably answer that users and needs are important.

But the reality is that each of us has an ego, and it is very difficult to let go of the ego. The knowledge we have learned and everything we have experienced will cause us to have a natural inertia for many things. We are used to the mouse, keyboard, API, and various potential rules, but in the AI ​​era, we may need to rethink whether these are really necessary, and whether the most effective way has changed to you directly handing yourself over to AI - AI knows everything you have seen, thought, and experienced, and then let it take you to the unknown .