Author: Alec Goh, Head of HTX Ventures

In the digital asset field, I have experienced multiple market cycles, most of which are driven by "narrative" and "technical advancements", and very few are truly game-changing.

I have gradually realized the combination of AI Agents and Web3 has the potential to do this over the past six months. However, the premise of all this is that the industry must go beyond the early stage of hype and truly build long-term value for end users.

We are already familiar with such stories. In 2017, the ICO boom far exceeded the implementation of products; in 2021, NFT broke out and entered the mainstream, and the infrastructure of most platforms has not yet been fully built. Both cycles have one thing in common - the bubble comes quickly and the callbacks are equally rapid.

The explosion of AI Agent tokens is exactly the same. Dozens of projects have been launched under the name of "Crypto + AI", quickly attracting a lot of capital and attention. Some projects have reached billions of dollars in a few weeks. But as before, it is difficult to build a sustainable ecosystem based on hype funds alone. Once the investment enthusiasm fades, user participation also declines, and most early projects have begun to lose their long-term relevance.

But in this round of fluctuations, some essential changes did occur.

In the fourth quarter of 2024, DeepSeek achieved a critical technological breakthrough, attracting the high attention of our HTX Ventures. Through reinforcement learning, this project trains large-scale models without relying on manual annotation of data. This means that the cost of developing autonomous intelligent agents has dropped significantly, giving AI Agents the possibility of large-scale and low-threshold deployment in Web3 for the first time.

This transformation is of great significance. If encryption wants to fulfill its promise of "decentralized, programmable financial infrastructure", intelligent agents with independent decision-making capabilities will become an indispensable underlying component. But at present, we are still quite far from this vision.

I am worried that the industry may once again only focus on catching up with the trend and forget to be down-to-earth. There are indeed a fewProspects have emerged, such as Eliza is building a composable Agent tool that adapts to multi-chain ecosystems such as Ethereum, Solana and TON; Launchpads such as Virtual and Clanker have accelerated token issuance; application products such as GRIFFAIN and NEUR are also trying asset management and governance tools.

But these are just early signs and have not formed a clearly verified system ecology.

Too many AI Agent projects are still speculative rather than sustainability as their goal. If this situation continues, we may face the old problems of capital flight, failure of expectations and slowing project progress again.

This will be a huge regret.

Because these application requirements are real. DeFi policies that are too complex for most users can be automated management through intelligent agents; on-chain governance can be analyzed by Agent, compare history, and voted according to user values; cross-chain automatic operations can also be directly executed by the agent based on user intentions.

These scenarios are no longer assumptions—but are real needs that have arrived. What the industry currently lacks is not imagination, but a stable execution, user retention ability and a feasible token economic model.

First, we must look at the AI ​​Agent project by the standards of examining the DeFi protocol.

This means breaking out of vague narratives, moving towards rigorous project logic, and raising key questions:

Who is the actual user?

Does it have indispensable practical value?

Where does income come from?

Will the protocol survive after the token incentive disappears?

The protocol should not only show technical potential, but also show real usage scenarios, stable revenue models and quantifiable influence. This is the fundamental difference between "proof of concept" and "worth building".

Second, in addition to applications, the industry also needs to build infrastructure to support the successful deployment of AI Agents, including the Agent framework, data layer, computing resources, etc. HTX Ventures will continue to pay close attention to the development of such infrastructure, as some of which will become a must-have in the future, just like EVM compatibility to the L1 public chain.

Finally, we need to build an industry culture that encourages iteration and transparency.

Not every project can be done right the first time, but teams that can persist in optimization and continuous polishing are likely to become the leader in the next stage of Web3.

The fusion of AI and encryption is not a "concept", but an "evolution". But if we want to truly seize this round of opportunities, the industry must shift from "focusing on attention" to "creating value."

This is not only the investment logic of HTX Ventures, but also the firm belief we have formed through years of experience and future judgment.

We have tools and we also have Builders. Now, what we need most is self-discipline.