Author: Wu Blockchain
This interview delved into the investment trends, market status, and future opportunities of the AI Agent project. Analyst defioasis shared his investment experience in on-chain assets, memecoins, and AI-related projects, including the rise of key projects (such as GOAT, Worldcoin, Turbo, Pippin, etc.) and the underlying market logic. Additionally, it analyzed investment methodologies, position management strategies, signs of market bubble cooling, and the importance of industry trend changes and project narratives.
Please briefly introduce yourself and review when you started researching AI Agents and when you officially started investing money, and what is the overall return rate so far?
defioasis:
Hello everyone, I am defioasis. I am usually interested in on-chain data, on-chain assets, and the derived gameplay. Since last year, my main focus has been on playing with on-chain things.
In fact, the research on AI Agents also stems from the exploration on the chain. I have been observing on-chain assets since last year, but pure Memecoins are not really my forte. I mainly just watch and am not particularly convinced to play with large amounts of capital.
The turning point was mainly at the end of October and the beginning of November last year, when GOAT surged to a market capitalization of several hundred million dollars in a very short period of time, and the listing of GOAT futures on Binance made me rethink this track. Of course, before talking about that, I still want to talk about some AI-related things before GOAT. Before GOAT, there were two relatively big events, one was Worldcoin, because the co-founder is Sam, the founder of OpenAI, so WLD was regarded as the Meme of OpenAI in Crypto and its FDV once exceeded $100 billion; the second was Turbo, which claimed to be an AI Meme created by GPT, and it also surged 200-300 times on CEX last year. So these two events show that people are very interested in and worth hyping the AI+Crypto theme.
Going back to the end of October last year, after GOAT was listed on Binance futures, I quickly associated it with Worldcoin and Turbo, and Binance has the ability to lead the track. At that time, there were 4 assets that I found very eye-catching, two of which were AI-related, one was ai16z and the other was ACT; the other two non-AI-related were LUCE and BAN. Now, ai16z seems to be much better than ACT, but at the time, the situation was that the founder of ai16z, Shaw, was still a nobody, and ai16z was launched by daosfun, and there were some issues with the token contract being mintable, and the volatility was particularly large because the pool was very small. At that time, I took a position in ai16z at a market capitalization of $25 million, spending about $3,000, but the pool was very small and it was once FUDed to a market capitalization of less than $10 million, so I didn't dare to add more positions. So I ultimately chose to overweight the more stable and larger-scale holder ACT. ACT can be considered the first position I officially took in AI investment. The first one was $3,000 of ACT bought on November 5th, and later I found it was also on Bitget, and I gradually bought more than $10,000 of ACT at a market capitalization of $22 million and an average price of $0.022. But at the time, I didn't even think that the position would be listed on Binance. I remember I was attending a conference in Bangkok, and I was actually quite shocked. After being listed on Binance, it reached a market capitalization of 7-8 billion dollars, but I actually sold most of it the next morning when it was at $500 million, and the rest haven't been sold yet.
Later, I also observed that AI Agents is the only track that has evolved from a general PumpFun to a vertical and scaled one, so I then delved deeper into this track for investment. After ACT, I also caught some other pretty good targets, such as Pippin, which I also heavily weighted and achieved more than 10x returns. The overall position for AI investment has reached 7-8x since November, and has recently pulled back a bit, but is still around 5x.
What is your research method for AI Agent projects? Can you give an example of the AI Agent project you have researched the most deeply, so that more people can understand your research methodology?
defioasis:
The current AI Agents or some AI-related assets are very different from the previous AI projects. Now it is basically whatever is based on the PumpFun-like fair launch to form the asset, so the project party or the founders may not have much asset, maybe even less than some snipers or big players. This means that the integrity, vision and background of the founders are very important, otherwise if they are not real-name or have no morals, they can abandon the project and start a new one anytime. So my methodology for AI Agents first starts from the people, whether they are really doing it, whether they have the ability to do it, and what they can achieve.
Let's take the example of the Pippin project. I first learned about Yohei Nakajima, the founder of Pippin, from the judges and guests of the Solana AI Hackathon on December 11th. At that time, I saw the judge list and found a Japanese, Yohei Nakajima, who is the founder of Pippin. At the time, I found his child-oriented AI Agents quite interesting, as I hadn't seen any similar positioning agents before. Another aspect is that as a judge, his "weight" is obviously higher than those who need to participate in the competition to prove themselves. Further research on Yohei Nakajima found that he is the founder of BabyAGI, which has more than 20,000 stars on its Github. At that time, I searched some information and found that BabyAGI is quite powerful, as an AGI concept product that has been cited by many media and papers, which also proves its strength. In addition, Yohei Nakajima is a partner of Untapped Capital, and the main Web3 project they have invested in is Pixel, which has been listed on Binance. Overall, Pippin's founder Yohei Nakajima is very powerful in both technology and capital resources, and as a real-name and reputable person, the probability of him rug pulling is much lower.
At that time, Pippin's market cap was around $20 million, which actually didn't attract much attention. A $20 million market cap also fits my requirement for buying assets, as I tend to buy in the $10-20-30M range. I gradually bought about 0.2% of the total, which is the maximum single position I define. It actually dropped below $10 million at one point, but I accumulated more at that time. The founder's background and technology won't change with the price fluctuations. So after accumulating, I just left it alone.
Later, Pippin announced that it would transform into a framework, from a single AI Agent to an AI Framework, and the valuation skyrocketed. Even though the framework hasn't been delivered yet, the market's recognition of the founder's technology and capabilities has pushed Pippin to surge to a $300 million market cap. Doing a framework means it may form a split-off model, and the market's pricing for frameworks is currently the highest in the AI agents track, and frameworks or ecosystems that can form a split-off model may have the opportunity to break through a market cap of over $1 billion.
Which AI Agents projects are you bullish on? What do these projects do? Why are you bullish on them?
defioasis:
I'm bullish on many, like Pippin, which I still hold, but the market cap is quite high now, so I won't talk about it. I usually choose around $20 million, but I actually don't heavily weight that many.
Currently, there are two main directions. One is to dig for gold in the Solana AI Hackathon, which has just ended, and there are quite a few award-winning projects that I'm currently screening. The other is to look at projects coming to Solana via Virtuals and collaborating with Jupiter, as Virtuals has already proven its success on Base. Of course, I'm also still looking at this area.
This mainly talks about some projects that came out of the Solana hackathon. I'll share one here, but this is not investment advice. A project I've been observing recently is AgentiPy, which is actually an open-source framework that uses Python to connect AI agents to Solana blockchain applications. According to the roadmap, they may launch an autonomous narrative trading bot in Q1 and a launchpad in Q2; the most important thing is that they mentioned the APY token will participate as a flywheel. The APY token economics are also well-designed, although it's also based on a fair launch like PumpFun, the team has put 40% of the tokens in Streamflow for a 2-year linear unlock, which shows the team's determination. The co-founder and CTO of AgentiPy have been followed by the Solana Official Twitter. Coming out of the Solana hackathon is at least an endorsement. Of course, it's still quite early and there's a lot of uncertainty.
From a broader perspective, I feel that AI is gradually moving towards the AI Application stage. In addition to frameworks, I'll also keep an eye on AI+ applications, especially AI+DeFi, which is the combination of AI's native narrative with DeFi assets and flywheels. This may present some good opportunities, but it's still in a very early stage and I haven't seen any good targets yet. I'm still just observing for now and haven't added any new assets recently.
What are your views on the current state of the AI Agents track and the market? Do you think the hype around AI Agents can last a long time or do you think the bubble has already peaked?
defioasis:
It has indeed been quite cool recently, but I don't think it will end yet. AI is still a fairly sensible thing, and AI outside the circle is still rapidly iterating and developing, in an upward trend in terms of technology and capital, which is a more important fundamental. In fact, many AI targets are actually driven by the outside world, whether in terms of narrative or talent. Shaw was once a marginal figure in Web2, and now the AI16z he founded has become a leader in Crypto AI, influenced by Shaw. I think more traditional industry tech talents will come to do AI-related things.
Crypto AI itself is significantly behind the outside world, so any wind and grass movement in the outside world, as well as major-level updates, will be transmitted to the Crypto to form new narratives and new sub-tracks.
From another perspective, from within the circle, AI Agents is the only vertical track that has formed scale from the general PumpFun. DeSci may be half of it, which has also been quite cool recently. Apart from that, there is no other track that can move from generalization to verticalization, which actually shows a strong demand for the AI narrative.
Now that AI Agents is cooler, it's a cooling of the previous overheating. The proliferation of individual Agents has forced everyone to do frameworks to compete, and people have become more fatigued by this kind of thing. So AI+ applications, especially AI+Crypto native narratives, if they can be established, I believe they will bring a new wave of hype and new opportunities.
Any insights or techniques on investment and trading, such as portfolio building and exit strategies?
defioasis:
The above is actually all about the investment targets, but I think position management is more important. The selection of targets is essentially about technology, resources and background, although it's based on PumpFun for a fair launch, the research approach is not much different from VC coins, with factors like technology, resources, background, team, endorsement, as well as analysis of the token structure, whale addresses, etc.
The focus is on position management. Nowadays, a reasonably good asset can generally go through three stages: PvP, second stage and listing. But most PvP projects are gone after the PvP stage. I mainly play the second stage, focusing on the $10-20-30 million market cap range, which I call the on-chain sweet spot. Because I found that some good assets will consolidate and fluctuate in this area after the initial surge and pullback. I usually only allocate assets in this range. I pay attention to those that have experienced 70% or more pullbacks and stabilized in the $10-20-30 million market cap range.
When I see a good asset, I'll add it to my watchlist, divided into S-tier and A-tier. S-tier are those that can form a pattern, like being a mother coin in an ecosystem, able to profit from the constantly generated child coins, or being a trading pair asset where the child coins create a wealth effect, constantly needing this mother coin for trading, similar to the PumpFun SOL and the mother coin flywheel of Virtuals. So S-tier are mostly framework-type, with the potential to form an ecosystem or pattern flywheel. That's why Pippin was able to take off, because the market had high expectations for it as a framework.
A-tier is mainly judged by narrative, background, technology, resources and team (without the ability to form a pattern). The background of the founders and founding team is very important, such as Solana Foundation background, lots of Github stars, having produced attention-grabbing products before, decent narrative with the potential to develop into a sub-track, etc.
Generally, I will build positions in both A-tier and S-tier, each time around $2-3,000. But I will be more strict in deciding whether to add more or overweight, I need to observe for a period of time, mainly looking at what the community and developers are doing daily.
I also have certain standards for the maximum single-coin holding. The maximum single-coin holding is 0.2%, i.e. around $4,000 for a $20 million market cap. Each time I decide whether to add more, I will re-evaluate the target, and if I find that it is inconsistent with the original purchase logic, I will decisively give up adding more, while keeping the rest of the position pending; if it falls sharply below the price range, I will also decisively give up adding more, while keeping the rest of the position pending. In principle, I will consider selling when it reaches 10x, around the $200-300 million market cap range.
Building positions in many targets is often to ensure I'm on the right track, adding positions cautiously and in batches, and setting a hard cap of 1.2% is to avoid being trapped in a single target due to excessive blind confidence. Of course, once I decide to add more, I will believe in my judgment of the target.
The above is about the second stage gameplay. I feel that the market, whether users, Devs or token factories, also seem to be more familiar with this second stage routine, and it seems to be quite difficult to play recently. So I'm also doing some lottery-style plays now, which is the PvP. At this time, I'm not limited to AI, I've prepared a few SOL, 0.1 SOL each time to buy in, watch the market, monitor and follow some selected addresses, to bet big with small amounts. From my experience in the last two days, in the current market, the lottery-style play is easier to profit. This may be a probability, with so many plates, as long as the input is small enough and diversified enough, and some high-probability addresses are used as auxiliary, the lottery-style can make up for all the previous losses as long as it hits once, which may be a good choice in the absence of a big bull market, and in a sense it's also preparing for the potential big bull market. Maintaining a sense of profitability and a positive mindset is extremely important.
