Will DeepSeek trigger a major reshuffle in the AI ​​Agent sector? Is it time to buy the dip or retreat?

Author: Azuma, Odaily Planet Daily

The AI ​​Agent sector continues to accelerate its correction today. Except for some memes of the DeepSeek concept, leading Framework/Launchpad/Agent tokens such as VIRTUAL, AI16Z, AIXBT, ARC, and other large and small tokens all recorded double-digit declines.

Although the decline is related to the overall weakening of the market, looking around the market discussion, some investors will attribute the collective decline of this round of AI Agents to the popularity of DeepSeek, believing that DeepSeek, as a new generation of more economical LLM, will have a positive impact on the current Crypto industry. The AI ​​landscape has caused shocks and even reshuffles.

I personally do not support this view.

The core point of the above doubts is actually - will the iterative development of LLM subvert the existing Agent? Putting aside whether DeepSeek's current reasoning quality is better than ChatGPT and Claude's arguments, assuming that DeepSeek has more promising iteration potential from the perspective of economic sustainability, will DeepSeek after the iteration create a new batch of Agents? ? Let these existing agents become "yesterday's flowers"?

To answer this question, we need to first understand the relationship between LLM and Agent. We tried to use DeepSeek itself and another "domestic light" Kimi to answer this question (to be honest, the quality of Kimi seems to be better in this question alone).

DeepSeek:

The relationship between AI Agent (artificial intelligence agent) and AI LMM (large language model) is that LMM can be used as one of the core components of Agent, especially when dealing with natural language understanding and generation tasks. AI Agent is an intelligent system that can autonomously perform tasks and interact with the environment, and LMM provides the Agent with powerful language processing capabilities, allowing it to better understand and generate human language, thereby enabling it to perform tasks in dialogue, recommendation, analysis and other scenarios. play a role in. In short, LMM is an important technical support for AI Agent to realize intelligent language interaction.

Kimi：

• Agent agents usually rely on LLM to understand and generate natural language. LLM provides powerful language processing capabilities, allowing the Agent to interact with users in natural language, understand the user's needs, and generate corresponding responses.

• Agent also not only relies on the language processing capabilities of LLM, but also combines many aspects of capabilities such as decision-making, problem solving, interacting with the external environment, and executing actions. LLM provides the agent with a basis for understanding and generating language, and the agent performs task planning and execution on this basis.

• Agents can also be combined with other tools and APIs to extend their functionality. For example, Agent can call search engines, database query tools, file reading and writing tools, etc. to complete more complex tasks. LLM provides language understanding and generation capabilities in this process, allowing AI agents to better interact with these tools.

To sum up, the operation of Agent requires the support of LLM, but the quality of its expression and interaction does not entirely depend on LLM. In fact, it is other capabilities besides LLM that determine the obvious differences between different Agents.

For example, the reason why aixbt is able to "crushing" other agents of the same type in terms of output is essentially because of its prompt word design, post-processing mechanism, context management, fine-tuning strategy, randomness control, external tool integration and user feedback. It has done a better job in terms of mechanisms and other aspects, so it can generate expressions that are more relevant to the industry - whether you call it a first-mover advantage or a moat, this is the current advantage of aixbt.

After understanding the logic of this relationship, let's now answer the core question in the previous article "Will the iterative development of LLM subvert the existing Agent?"

The answer is no, because Agent can easily inherit the capabilities of the new generation of LLM through API integration to achieve evolution, thereby improving interaction quality, improving efficiency, and expanding application scenarios ... especially considering that DeepSeek itself provides a solution that is compatible with OpenAI API format.

In fact, the Agent that responds quickly enough has already completed the integration of DeepSeek . ai16z founder Shaw said this morning that Eliza, the AI ​​Agent construction framework developed by ai16z DAO, completed supporting DeepSeek two weeks ago.

Under the current trend, we can rationally assume that following ai16z 's Eliza, other major frameworks and agents will also complete the integration of DeepSeek as soon as possible. In this way, even if there will be impact from some new generation of DeepSeek Agents in the short term, in the long term the competition between agents will still depend on the external capabilities mentioned above, and at this time the development results brought by the first-mover advantage Accumulation will show up again.

Finally, let’s post some comments from big guys about DeepSeek to recharge the faith of the adherents of the AI ​​Agent sector.

DeGods founder Frank said yesterday: “ People’s idea of ​​​​this matter (DeepSeek iterating the old market) is wrong. Current AI projects will benefit from new models like DeepSeek. They only need to replace OpenAI API calls with DeepSeek. Output will improve overnight. The new model will not disrupt agents but accelerate their development. "

Daniele, a trader who focuses on the AI ​​sector, said: “If you are selling AI tokens because the DeepSeek model is cheap and open source, then you need to know that DeepSeek is actually very helpful in extending AI applications to digital data at low-entry pricing. Millions of users. It’s probably the best thing that’s ever happened to the industry. ”

Shaw also published a long article this morning in response to the impact of DeepSeek. The first sentence of the article is as follows: " A more powerful model is always a good thing for the Agent. Over the years, major AI laboratories have been surpassing each other. Sometimes Google is ahead. , sometimes it 's OpenAI, sometimes it's Claude, and today it's DeepSeek... "