Author: Foxi (DeFi / AI)
Compiled by: TechFlow
"Agents are not just here for your Twitter feed - they are reshaping the entire digital economy."
This article will delve into how to discover the next leading AI agent (Alpha) by understanding the methodologies. Remember: once everyone sees it, it has already lost its lead advantage. The key is to be two steps ahead, and by applying this methodology, you can seize the opportunity and succeed.
Special thanks to the AI report from Delphi_Digital and the contributions of other pioneering researchers.
The Golden Rule: Consensus
In the market, everyone tries to sell their assets to others at a higher price. Your goal is to exit at the right time, not to be overly attached to any asset. To successfully exit, you need to find a project with "consensus". This means that most people in the market recognize the value of the project, providing liquidity support for you. Building consensus is a complex topic, which we will not delve into today.
How to Make Money?
- Predict an emerging trend (consensus) and hold a firm belief (knowledge is needed to support your judgment).
- Invest in potential projects that are innovative and legitimate.
- Sell your assets when the market forms a consensus.
This article will focus on the most challenging step: prediction. If you don't have a clear understanding or belief about the trend, it's difficult to independently judge the value of potential projects. In this case, you either rely entirely on luck or wait for others' cues, and ultimately become the object of others' profits. Let's get started.
Part 1: The Three Waves of Agent Evolution
The year 2024 is marked by OpenAI's o1 system and a valuation of billions of dollars, while 2025 is considered a pivotal year for the financialization of artificial intelligence. However, to find true value in this rapidly evolving field, we need to go beyond the hype of "AI agents" and delve into the fundamental changes behind them.
Understanding which stage of the technology adoption cycle we are currently in is crucial for seizing opportunities. The development of the agent economy can be divided into three stages:
Image: From Foxi (DeFi / AI), compiled by TechFlow
Wave 1: Human to Agent (Current Stage)
We are currently in this stage, with typical examples being simple chatbots on X (@aixbt_agent). These agents mainly serve as research assistants or execution tools for humans, helping to complete simple tasks. Although they have some value, they do not require disruptive infrastructure reforms. Most current agents are similar to a customized ChatGPT, helping users with simple research tasks, but with limited autonomy. They cannot independently manage resources, take risks, or pay service fees.
Wave 2: Agent to Human (Emerging)
Agents in this stage are beginning to truly demonstrate their potential. They can independently complete daily tasks, such as running trading strategies, optimizing home energy use, and even negotiating and paying bills. These tasks no longer require frequent user intervention. Tools like Stripe's Agent SDK already cover some scenarios, and also reveal a trend: the traditional fixed monthly or annual fee model will gradually be replaced by a pay-as-you-go pricing model.
As agents take on more responsibilities, they need to pay for computing resources, API query fees, model inference costs, and other expenses - these costs may be real-time and on-demand. However, the limitations of existing payment systems are exposed in these scenarios, as they are not designed for real-time micro-payments. Cryptocurrencies have advantages in this area, providing faster settlement speeds, lower transaction costs, and more flexible payment methods to better meet these new needs.
Some on-chain real-world cases include:
- Automated trading systems
- Yield optimization tools
- Portfolio rebalancing mechanisms
Wave 3: Agent to Agent / Collaboration (Future)
This is the most promising stage of the agent economy. We have already seen some early experiments in agent-to-agent commerce, such as projects like Terminal of Truth and Zerebro. However, the true potential of these explorations goes far beyond the applications of social media tokens:
- Resource markets: Computing agents can negotiate with storage agents to optimize data storage locations, improving resource allocation.
- Service optimization: Database agents can negotiate query optimization services with computing agents to improve efficiency.
- Financial services: Risk assessment agents can trade with insurance coverage agents to provide smarter protection plans for users.
This stage requires infrastructure specifically designed for machine-to-machine interactions. Traditional monetary systems are human-centric, emphasizing human verification and control, but lack the flexibility needed in an economy dominated by autonomous agents. Stablecoins, with their programmability, cross-border payment capabilities, fast transaction settlement, and support for micro-payments, become key tools in this domain (yes, I'm also very optimistic about the stablecoin space).
Part 2: Analyzing the AI Stack in Web3
Delphi Digital has proposed a clear "Decentralized AI" stack classification framework. The framework contains 6 key directions (which I believe have great potential), each with several sub-domains that can be viewed as narratives or innovation directions. I believe that in the coming months, there will be a leading project in each sub-domain with a market capitalization of at least $500 million.
6 Key Directions and Their Prominent Examples
- Applications - aixbt
- Agent Enablement / Coordination - Virtuals Protocol
- Privacy Preservation - Phala Network
- AI Training / Inference - Ritual
- Compute Services - io.net
- Data Storage - Arweave
Image: From Foxi (DeFi / AI), compiled by TechFlow
Whether you save this article or this image, it can serve as an important reference guide as you explore new narratives.
Each direction contains many sub-domains, so this article will not go into them one by one. However, I expect the capital inflow into these directions to follow a certain "order", depending on the market's maturity and development stage. Here is a brief visual illustration:
Image: From Foxi (DeFi / AI), compiled by TechFlow
Why choose aixbt? It's just because it's interesting, not because it will become the ultimate champion of the AI industry.
Stage 1:
We can see that a large number of agents are being launched every day, most of which are just simple chatbots or ChatGPT encapsulations with limited practical use, but their meme properties make them appear interesting. As we know, the marketability of a product is not directly related to its practicality. Therefore, interesting agents with enough imagination tend to be pursued by the market.
In this stage, agent enablement platforms (such as Launchpad) become the underlying support layer for agents.
Example: Agents generate revenue through interaction with users, and this revenue is used to repurchase and burn the tokens paired with $Virtual. This mechanism directly links the success of the agents to the platform value, aligning the incentives across the entire ecosystem.
Image: The current market focus has been highlighted with a red frame.
Stage 2:
As "meaningless content" posted by chatbots on Twitter becomes ubiquitous, this narrative has gradually lost its appeal. As a result, we are seeing more and more people incorporating the concept of "privacy" into intelligent agent applications (such as @aipool_tee and @sporedotfun). Interestingly, most people do not understand complex technical terms like TEE (Trusted Execution Environment), FHE (Fully Homomorphic Encryption), and ZKP (Zero-Knowledge Proof). This makes intelligent agent applications appear more innovative, even if they may not actually implement these technologies. Nevertheless, the inclusion of these concepts has indeed enhanced the "practicality" of intelligent agent applications.
The next development direction for intelligent agents is to enter the DeFi and wallet domains. In the future, we will see intelligent agents helping users with token swaps, cross-chain bridge operations, optimizing transaction paths, and reducing transaction fees. These functionalities will be directly integrated into the user interface of wallets. However, unlike the first stage, these intelligent agents in this stage need to compete on "technical capabilities" rather than relying on "belief hype". As a result, some tokens (such as $BUZZ and $ACOLYT) are being pursued by the market due to their strong AI teams or developers behind them, even though their technical level may still lag behind Web2 AI experts.
As the intelligent agent economy continues to evolve, the "wealth effect of AI intelligent agents" will attract a large number of top-tier Web2 AI developers to enter the Web3 field in search of new opportunities. This will spawn many powerful AI projects. If you want to seize these opportunities, I suggest you pay more attention to LinkedIn, rather than just staring at Dexscreener.
Stage 3:
Intelligent agents will ultimately mature, with their core value coming from AI reasoning, data processing, and distributed computing. We will see the emergence of the following key infrastructure:
- TEE-based secure key management systems
- Dedicated data availability (DA) layers for storing and retrieving context of large language models (LLMs)
- On-chain Oracles for providing trusted data sources
- zkVM frameworks for enabling verifiable computation execution
- Chain abstraction solutions for simplifying cross-chain interactions
This infrastructure will support large-scale, trustless computational resources while ensuring the security and verifiability of interactions, data flows, and outputs. At that time, advanced infrastructure projects like Ritual, io.net, and Story Protocol will transition from mere speculative objects to become core tools driving the next generation of AI innovation.
Part Three: So what should I invest in now?
Based on the current market situation, we are transitioning from Stage 1 to Stage 2.
If intelligent agents can "proactively" provide value to users, rather than just posting on Twitter or analyzing token prices (which GPT can already do), they will be more appealing. I hope intelligent agents can autonomously manage resources and make decisions on behalf of users, and even independently complete transaction settlements. Here are some more complex application directions that I personally would consider investing in:
- Automated trading systems
- Portfolio rebalancing tools
- Virtual reality technologies
- AI-driven smart contracts (such as prediction market arbitration)
- Other innovative applications
Of course, the above directions may still be too broad for most people (including myself). According to the "Consensus Golden Rule", I observe that the current market consensus is mainly focused on the following:
I. Shaw
He is the most influential belief leader in the current AI super-cycle, just like Murad in the memecoin super-cycle. Do not ignore the projects promoted by ai16z and Shaw or other core ai16z members. Any project they promote is likely to be a buying opportunity.
Note: Be careful of those who claim to be "ai16z partners", as they are sometimes not genuine core members. In fact, I could also randomly call myself a "partner".
II. Solana Hackathon Award-Winning Projects
The Solana ecosystem is actively promoting the development of some innovative projects. To seize the market first-mover advantage, you need to identify high-quality projects in advance. I have already screened some projects for you, and you can refer to my posts. Although I am not an insider, I have evaluated these projects based on my personal experience. Of course, you should also conduct your own in-depth research.
Link: Google Sheets link
To truly profit from these narratives, you need to leverage some tools, or you may miss the best investment opportunities. For example, @AgentiPy is one of the high-quality project lists I have, and it initially did not issue any tokens. Once it launched its token, the market cap quickly soared to $40 million, but I personally missed the opportunity due to the lack of timely tool support. Therefore, I suggest using tools to monitor project dynamics, such as tracking their Twitter, and being able to intervene immediately when the tool detects the release of a contract address (CA). More tutorials on these tools will be released soon.
III. AI Frameworks
You may have noticed that there are more and more infrastructure projects in the blockchain industry, and even many dApps are starting to transform into Appchains. The reason behind this is simple - infrastructure projects are usually able to obtain higher valuation space. However, not all frameworks have equal value. You can choose the market-leading projects, or delve into the technical documents and GitHub to find infrastructure projects with strong technical capabilities.
If you find this research work too complex, a simple strategy is to focus on high-quality tokens like $ai16z or $Virtual, and find the right entry point. Do not invest your SOL or ETH in those AI scam projects without any technical support.
IV. DeFi x AI
As mentioned earlier, I am very optimistic about the prospects of the DeFi and AI integration narrative (DEFAI). This trend is in line with my prediction for Stage 2, where the interaction model between intelligent agents and humans is gradually maturing. A good starting point for research is to delve into some DeFi x AI projects and their sub-sectors, understanding their specific application scenarios and technical implementations.
Finally, understanding these market dynamics is crucial to grasping the real opportunities in the intelligent agent economy. Although the current market is still dominated by simple social media intelligent agents, the true value lies in the infrastructure and frameworks that can drive the next generation of autonomous economic activities. I will continue to hold $ai16z and $Virtual, as I believe they have tremendous potential in the future.
