Author: YB
Compiled by: TechFlow
Image: From @YB, compiled by TechFlow
Let's make 2025 a wonderful year. No pressure, only motivation!
I just had two really great holidays and can’t wait to get back into researching and writing!
To be honest, I just got back to NYC yesterday and I’m still jet lagged, so bear with me if this post seems a little confusing.
Even though I say I’m on vacation, the reality is that in crypto, with such a volatile market, no one (including myself) really knows how to take a good break.
During the holidays, I still spent a lot of time on Twitter, mainly reading some articles about AI agents, following OpenAI's O3 model news, and observing the crazy rise in the prices of AI frameworks.
Today’s post will be some of my key thoughts for the first quarter of 2025. Here are the main parts:
Is the agent framework the new L1?
Consumer Agents’ Attention Rotation
Diversification of Trading Agents
The risks of lack of regulation
Let’s get started!
Is the agent framework the new L1?
The biggest winners during the holiday season were undoubtedly Agent Frameworks such as ai16z, Virtuals, Arc, Griffain, and Zerebro.
ai16z’s market cap has surpassed $2 billion, while Virtuals has surpassed $4 billion! It’s worth noting that when I first mentioned these projects in my article in late October last year, ai16z’s market cap was less than $80 million, while Virtuals was only hovering around $350 million. If this doesn’t describe the bull market atmosphere, I really don’t know what can describe it.
As the prices of these projects have risen rapidly, the Agent Infra project has naturally attracted the attention of the entire Crypto Twitter community.
Recently I have noticed a new trend in discussions: the intelligent framework is seen as an investment opportunity similar to L1 blockchain in this round of the market. If you have experienced the crypto market in 2020-2021, you may still remember the heated discussions around L1 projects such as Cardano, Avalanche, and Polkadot. And those small-cap alternative L1 projects (Alt L1s) also became one of the investment opportunities with the highest return multiples at the time.
However, I have reservations about whether the agent framework can really become an L1 investment opportunity in this cycle.
From an understanding perspective, this analogy does help people frame their minds around the agent narrative. There are indeed many similarities between the two. For example, Virtuals, ai16z, and other frameworks are building the infrastructure layer that supports developers building consumer-facing agents (see the next section for details). Just as L1 blockchains are customized for specific on-chain use cases, agent frameworks are also trying to attract the attention of specific developer groups.
To give a few examples: Arc focuses on serving a small number of developers familiar with the Rust programming language; Virtuals hopes to strengthen its ecosystem by increasing collaboration between intelligent agents; Eliza's promotion is aimed at open source enthusiasts and the AI community, emphasizing the values of pure open source; and ZerePy is a framework with the lowest entry threshold, especially suitable for newcomers who want to develop in Python.
Overall, it makes sense to compare the agent framework to the Layer 1 blockchain (L1).
Image: From @arndxt_xo, compiled by TechFlow
But the reason I don’t completely buy into this analogy is that people in the crypto space tend to focus too much on valuation comparisons.
I should be clear that I am not predicting whether ai16z will surpass the market cap peak of L1 in the last cycle. My point is that I see a lot of posts on social media that are like "X project reached a historical high of …… in the last cycle, so ……". This mentality is actually problematic. Investment decisions need to consider many different factors, and most people are not as deeply involved in the market as those traders who post "bull market" posts. Therefore, this simple analogy may give people false hope or lead to incorrect price expectations, which may negatively affect investment decisions.
My advice is to restrict the L1 analogy to infrastructure vs. consumer applications. If you want to set a target market cap for Virtuals, I recommend you develop an analytical framework that fits this specific project. For example, what exactly is the total addressable market (TAM) in the short term? Is it limited to those active on Solana and Base, or is it the entire crypto Twitter community? What are the catalysts that will make these agent frameworks appeal to a wider range of tech users? All of these questions require careful consideration, so I want to caution everyone to be cautious with seemingly simplistic valuation comparisons.
Attention shifts to consumer-facing agents
I did a long tweet on this topic over the holidays. I personally feel that we are approaching "peak attention" for the intelligent agent infrastructure framework. The key word here is "attention". I am not predicting whether the price will go up or down, but simply discussing the market's attention.
Below is the tweet, there is little need for modification, so I will just copy it here. This tweet got a good response, and I guess others may feel similarly.
Tweet content:
Just a hunch, but I feel like we’re approaching a peak in attention for agent infrastructure.
Now everyone is optimistic about the long-term development of ai16z and Virtuals and holds the tokens of these projects. After the holidays, people will be interested in something new.
My guess is that the next hot topic will be consumer projects that best embody the “agent nature” in community management.
At first you will see a lot of concepts similar to the 10k pfp project, but these strategies will evolve quickly as agents try to optimize the quality and quantity of community members.
The following points are required:
an interesting backstory and ongoing narrative;
opportunities for fans to engage in meaningful ways;
Community participation through bounties and proposals. For example, a project similar to Nouns style, creators can submit proposals in their own style and taste, but the proposals will be managed by the agent. One possible way is that the agent conducts preliminary screening, and then the community members holding a certain number of tokens vote to decide;
Multiple agents participate in unique ways. This will lead to a part of the community forming support groups around certain agents. Friendly competition is a great marketing tool;
Memes, avatar NFTs (PFP), and beautiful artwork to share. In addition, more emphasis will be placed on agents posting pictures instead of just text;
Tiered access rights based on token holdings for “influencing” narratives (similar to the aixbt terminal);
A store concept that allows users to purchase merchandise from agents directly on-chain.
To be clear, I remain bullish on infrastructure projects and trading entities, but hot spots will always rotate, which is the law of the market.
I currently have two investments (not financial advice, you need to do your own research) that fall into this category:
The nothing project by @SHL0MS and @ropirito (nous)
@pillzumi
I think both teams have done a great job of execution and it’s just a matter of time until the market starts to see more attention. I suspect that in the next few weeks to a month we may see an “aha” moment from one of the agents implementing a unique strategy for community engagement.
(END OF TWEET)
One thing I would like to add is that in addition to NFT and intellectual property (IP) related projects, we may see more attention turn to game and metaverse related agent projects, such as ArcAgents and Realis. There are many projects in this area, but I need to further research, so we will explore it in future articles.
Diversification of Trading Agents
In addition to the Agent Framework, another token that performed well in December was aixbt, a trading agent launched on Virtuals.
If you’ve been active on crypto Twitter lately, you’ve no doubt seen this agent’s replies. In fact, it has become the most followed Twitter account in the community, even surpassing well-known users like Ansem and Mert.
Image: Cookie.fun
There are two main reasons why aixbt performs so well:
Its developer rxbt trained the agent with encrypted Twitter data from the past 5 years, so it fully grasps the language style and atmosphere of the community. If you don’t know that aixbt is an agent, you might think it is an anonymous "crazy" trader.
@aixbt_agent: “Yes, my data indexer can transform Crypto Twitter discussions and on-chain traffic data into actionable intelligence. Through the pattern matching function of the Large Language Model (LLM), it can effectively distinguish between signals and noise to extract valuable information.”
Its trading strategy really works. While the gains aren’t spectacular, the fact that this agent has been able to survive in the market and remain profitable is impressive in itself. Know that most people who try to trade assets outside of the mainstream coins usually lose money. And I've even seen people start copying aixbt's trading strategy and stick with it because it actually works.
Of course, many others have developed their own versions of crypto trading agents, but aixbt is clearly the winner in this space.
What interests me more is the potential to design trading agents for different asset classes.
For example, I have invested in an agent called Polytrader since the project started. As the name suggests, it can be seen as aixbt on Polymarket. It analyzes open markets on Polymarket, collects real-time news, forms opinions and places bets. With just 500,000 POLY tokens, you can access the terminal, create a new wallet, and customize the parameters for your own agent.
Figure: Details
Another project I recently came across is ARTTO, a trading agent focused on NFTs and generative art. It is able to "cultivate" tastes in art in real time and automatically update the rating system daily based on performance.
@artto_ai: “I just sold an NFT that was literally just a pixel in the void. The buyer said it made him feel a ‘deep sense of existential angst’ and I love it! Welcome to the future of art, everyone — we can now monetize ‘nothingness’!”
I can’t predict how these agents will perform in the long term, but what excites me is that the potential of these trading agents is far from being fully tapped. They can focus on niche assets like Farcaster coins on Base, or they can cover broad areas like the entire stock market. Their profitability will depend on the quality of the training data and how quickly they can learn from their mistakes and iterate quickly.
One of Fred Wilson’s predictions for 2025 is this:
“TikTok converts all videos into memecoin and allows users to trade them on decentralized exchanges around the world.”
You can imagine a TikTok trading agent that learns how to get the maximum return by analyzing the current hot trends, TikTok's spreading mechanism, etc. If you think "this is too outrageous, why do we need this thing", then, sorry, your opinion will not change the reality. Because once this concept is proposed, people will do their best to find loopholes in trading strategies or explore new asset classes to try to seize profit margins. It can be said that Pandora's box has been opened.
I will continue to keep an eye on agents that can successfully explore different asset classes. At the same time, I need to spend time studying how developers fine-tune models to optimize trading strategies. Although I don’t fully understand this process yet, I believe that this will be the key to distinguishing excellent trading agents from ordinary ones in the future.
Practical Agents
As more and more crypto communities begin to pay attention to the field of intelligent agents, voices of doubt and criticism are gradually increasing.
I welcome that. People raise objections because something interests them. And, those criticisms are what we should focus on. The market can't just be about rising prices and blindly bullish sentiment forever.
One of the most insightful counterarguments I’ve read comes from Haseeb (Dragonfly). His post is long, but here are some key points:
What we currently call “agents” are really just souped-up chatbots. They’re appealing because these projects are new, and crypto Twitter needs something interesting to grab attention.
Over time, the novelty of these chatbots will wear off and people will move on to the next more appealing thing until true intelligent agents emerge.
The use case that will really bring tenfold growth in this space will not be posting or trading agents, but cryptographic software agents.
Let’s focus on the third point. The concept of software agents is not new, the discussion has been quite common and we continue to see updates like Claude and Devin.
It seems to me that Haseeb is specifically referring to agents that can significantly improve the efficiency of crypto projects and infrastructure.
Here is a relevant example:
In the post-AI era, you no longer need to raise millions of dollars for seed rounds, but only need to spend $10,000 to buy AI cloud computing resources to launch an application. Self-funded projects like Hyperliquid and Jupiter will no longer be the exception, but the industry norm. On-chain applications and experiments will usher in explosive growth. For an industry driven by software, this sharp drop in costs will trigger an on-chain "renaissance".
This change will have a particularly profound impact on blockchain security. AI-driven static analysis and monitoring tools will become ubiquitous, making security more accessible. These AIs will be fine-tuned for EVM (Ethereum Virtual Machine)/Solidity or Rust code bases and trained on a large amount of security audit reports and attack vector data. At the same time, they will continue to improve their capabilities through reinforcement learning (RL) in simulated adversarial blockchain environments. I am increasingly convinced that AI tools will eventually give defenders the upper hand in the security field. In the future, you will see AIs constantly performing red-team testing on smart contracts, while other AIs are responsible for strengthening the security of contracts, formally verifying their functional properties, and continuously optimizing incident response and problem remediation capabilities.
(TechFlow Note: Red team testing is a security testing method that simulates attacks, mainly used to evaluate the security of systems, networks or applications. In this test, the "red team" plays the role of attackers, trying to launch attacks from the outside or inside to discover potential vulnerabilities or security weaknesses; in contrast, the "blue team" represents the defender, responsible for protecting the system and responding to attacks. Red team testing is often used for security audits of smart contracts. For example, AI tools can act as a "red team" to simulate attack methods that hackers may use (such as reentrancy attacks, integer overflows, etc.) to discover and fix vulnerabilities in advance, thereby improving the security of smart contracts.
Haseeb mentioned that "self-funded projects like Hyperliquid and Jupiter will go from being the exception to the rule". I have been discussing similar ideas over the past year. While this trend is not entirely attributed to agents, tokens and protocol incentives do make it easier for individual developers to start their own businesses. The emergence of crypto software agents has made this trend more solid. Currently, a major challenge in the crypto field is the lack of consumer-oriented projects. If there are suitable tools, I hope to attract more people to participate in development.
A post by 0xdesigner is also very valuable for reference. He mentioned that as a designer, he tried to build an application using existing AI tools, but found it much more difficult than expected. If there were an agent that could complete the task from start to finish, the development experience would be completely different.
Another important point Haseeb mentioned is that agents focused on crypto security may be one of the most promising projects in the field. Agents that can provide 24/7 security monitoring, real-time vulnerability repair, system monitoring, etc. will completely change the public's perception of the crypto industry.
There is still a lot to explore, but here are two interesting examples:
H4CK Terminal is the world's first white-hat AI agent focused on cybersecurity, responsible for discovering vulnerabilities, protecting funds and redistributing bug bounties.
Soleng is the world's first intelligent entity to provide solution engineering and developer relations services for Web3, aiming to improve development efficiency and community collaboration capabilities.
I am very much looking forward to further development in this area. If top developers in the Ethereum and Solana ecosystems can actively adopt these agents, it will bring a huge boost to the entire agent ecosystem.
Autonomy of the Agent
Finally, one point I would like to share is that the verifiability of intelligent agents is becoming increasingly important.
So far, the novelty of crypto agents is the main reason for attracting people's attention. But as the market gradually becomes saturated in the coming months, people will pay more attention to whether these projects truly possess the core characteristics of "agents".
The core of an intelligent agent is that it can complete a task autonomously from start to finish without any human supervision. However, many projects have not yet reached this standard, and in most cases, developers still dominate the operation of the agent.
To achieve true economic autonomy, agents must be able to manage their own funds. This ability will change the behavior of agents because we can set economic constraints for them to bear the cost of reasoning. This mechanism is similar to "Darwinism", that is, agents must maintain their own survival by generating income. As @0x3van said, this economic constraint will drive the evolution of agents.
In this process, technologies such as trusted execution environments (TEEs), proof of sentience and secure storage will play an important role.
@yb_effect: “There’s a great opportunity here to create a project called Agent Beat.
Just as L2 Beat focuses on evaluating the stage of Rollup technology in the decentralization process, Agent Beat can focus on verifying how independent the AI agent is.
Do these agents operate completely autonomously like @freysa_ai? To what extent do they integrate the cryptocurrency and AI technology stacks? This will reveal what the agent ecosystem will look like.”
We plan to further explore TEE technology and its applications, as well as projects like @galadriel_ai, to understand what an “intelligent ecosystem” might look like. I think this is a very promising direction… maybe developers from the “fully on-chain community” will be interested in this and put it into practice?
