Written by: cryptoHowe.eth

Recently, ArkStream Capital published a research report on the AI ​​Agent track. After reading it, I think it is quite pertinent and I agree with many of the views. This time, I will extend some of the points mentioned in the article and talk about my views and opinions. I also welcome everyone to communicate with me.

Disclaimer: This article is highly subjective. The views mentioned in the article do not constitute any investment advice. They are only for communication and sharing. At the same time, this article is based on personal cognition and existing data, and will be updated at any time in the future.

Why does the AI ​​Agent track occupy such a large market share?

From the report, we can see that AI Agent currently occupies nearly a quarter of the market share of the entire AI track. I personally believe that there are two reasons why it can occupy such a large share:

1. Agent has a wide range of application scenarios, low barriers to entry, and short product cycles

Currently, the AI ​​track is mainly composed of five aspects: data, storage, computing, algorithms, and communications. In terms of data, sufficient resource accumulation is required and it is easily affected by geopolitics. Storage and computing are currently very demanding and resource-intensive, while algorithms and communications have very high technical barriers.

The Agent track is in a position where "everything is just right". It does not need to be supported by massive data, storage and computing power like general large models, nor does it need to make major improvements in algorithms and communications like squeezing toothpaste. As long as it can meet the development needs of the product itself, it will be fine. Therefore, compared with other AI products, Agent products do not have a very high threshold, have a wide range of usage scenarios, and have a relatively short overall development cycle. The project starts quickly, commonly known as "small and beautiful".

2. Agents are closer to the needs of the general public, easier to implement, and meet the narrative of Mass Adoption

The narrative of Agent products is actually very similar to the chain abstraction track, which is to allow users to focus only on their own needs without considering the implementation path and various participants in the middle. Agent has undoubtedly played a huge role in helping Web2 users enter Web3. Users no longer need to learn basic knowledge such as wallets and signatures from scratch, but can directly express their needs through natural language to automatically implement related operations. If a user wants to exchange all BTC for ETH, then the Agent will automatically plan the relevant interaction process and automatically perform cross-chain, transaction and other operations, and the user only needs to wait for the Agent operation to complete to get the result he wants. Therefore, Agent can be classified as one of the directions that can truly realize Mass Adoption.

The survival dilemma of content generation agents

The report also divides Agent products into infrastructure and content generation. Most products currently belong to the infrastructure category. Why is the development of content generation relatively slow? In other words, what is the survival dilemma of content generation Agents? I think there are two main reasons:

1. Content generation is more of a way to satisfy emotional needs, but emotional needs are difficult to price.

To put it simply, the business model is difficult to close. For infrastructure products, it is only necessary to provide relevant services or resources, such as providing relevant computing power or model services to AI developers. In this process, the price of the product is easy to measure, that is, what type of graphics card computing power the user used and how long it took, and the relevant price can be obtained through elementary school arithmetic, and these prices fluctuate little.

For content-generating products, it is very difficult to satisfy users’ emotional needs in a sustainable way. This is because users’ emotional needs are unstable. They may be very happy today, but suddenly become depressed tomorrow. Users’ willingness to participate in product interactions is also different. Different users have different emotional needs at different stages, and their willingness and degree of payment are also different. Therefore, prices fluctuate greatly.

2. How to determine whether the Agent's content meets the user's needs is a big problem

In content generation products, human subjective consciousness occupies a large part of the product. For example, there is no standard to measure whether the effect of image generation is satisfactory, and it depends more on the user's personal feeling, unlike the computing power market, which has clear standards and market prices. Therefore, it will be more difficult to retain and convert users of such products.

My personal opinion on AI Agent

Regarding the future development of the AI ​​Agent track, I personally think there are four points that need to be noted:

1. It is difficult for pure agent narratives to gain a competitive advantage in the market, and differentiated competition is needed. In the current environment, more and more AI projects will use agent as one of their narratives, and it is difficult for pure agent projects to stand out. Imagine that among hundreds or thousands of AI projects, it is difficult for only the agent narrative to attract users' attention. After all, good wine will be hard to sell if it is not well known.

2. AI Agents will gradually transform from being independent to interconnected AgentFi. Currently, products in the Agent track are independent of each other, and their data or services are not interoperable. Users need to provide relevant personal data from scratch when using different Agent products. If there is a reasonable way to connect different products, that is, users can use the Agent trained in product A on product B, the imagination space and user experience will be better.

3. Projects that sell water logic will be the first to come out and occupy the majority of the track market. Simply put, everyone is developing agent products, so I will make a tool that can efficiently develop agents. Products of this type that can come out are basically gold shovel projects that are sure to make money.

4. The revenue of Agent products mainly comes from toB, while toC is more of a strategy to accumulate word-of-mouth. This should be considered a common phenomenon in the AI ​​track. The willingness and ability of C-end users to pay are far less than those of B-end users. Therefore, if a product wants to make a real profit, it depends more on the quality of the B-end partner. However, the publicity and promotion ability of C-end users should not be underestimated. Having enough users to use the product can be a good help for the subsequent publicity and promotion of the product.

Finally, here is a summary of a good AI Agent framework that I saw recently.