Author: Kevin, the Researcher at BlockBooster
The term AI agents comes from OpenAI's roadmap. Sam Altman divides the capabilities that AI should have into five parts, with the third step being AI agents, which will be frequently encountered in the coming years.
What AI agents can do is autonomous learning, decision-making, and task execution. Of course, based on intelligence and capability, Stuart Russell and Peter Norvig in their book (Artificial Intelligence: A Modern Approach) categorize AI agents into five directions:
Simple Reflex Agents: React only to the current state.
Model-Based Reflex Agents: Consider historical states during decision-making.
Goal-Based Agents: Focus on planning and finding the best path to achieve specific goals.
Utility-Based Agents: Aimed at weighing benefits against risks to maximize utility.
Learning Agents: Continuously learn and improve through experience.
So, what level do the AI agents appearing in the current market or industry belong to? What direction are they heading?
OpenAI o1 has reached Level 2 AI. Personally, I believe that current AI agents in the industry are between Level 2 and Level 3, specifically at Level 2.5. This does not mean that agents in the industry have surpassed OpenAI; in fact, web3 agents are still at the stage of GPT wrappers. So why Level 2.5? Because, through human or programmatic intervention, let's call it mediation, a combination of the GPT wrapper and mediation forms a shape that is hard to scrutinize but possesses objective proactivity. It is an extension of a certain direction of OpenAI's model application. Considering what agents can do, they are the most basic simple reflex agents. Some of these agents may consider historical states, but they require active input. Only by continuously feeding data can agents complete learning, which is a passive model training method and far from reaching the state defined by Level 3. The last three types—Goal-Based, Utility-Based, and Learning Agents—have not yet entered the market. Therefore, I believe that current AI agents are still in the early stages, undergoing fine-tuning of Level 2 general LLMs, and have not structurally departed from Level 2. So, can evolution to Level 3 be achieved solely by crypto? Or do we need to wait for companies like OpenAI to develop it?
Why discuss the potential of Base or Solana to become the narrative center for AI agents?
Before discussing which ecosystems can promote the birth of Level 3 agents, we should identify which ecosystem has the potential to become fertile ground for AI agents. Is it Base? Or Solana?
To answer this question, let's first review how AI has affected Web3 over the past two years. When OpenAI first released ChatGPT, the protocols in the industry were still following habitual thinking, quickly flooding into the infrastructure bubble. Many computing power/inference aggregation platforms appeared, along with AI + DePIN infrastructure. The commonality between the two is that they build grand visions; this does not mean that grand visions are bad. In fact, agents can also build such visions, but in terms of implementation and user needs, such large infrastructure protocols have not considered things comprehensively. Because the market demand they want to pull up is far from saturated in the traditional internet industry, user education and market education are insufficient. Under the impact of the Memecoin craze, the AI infrastructure of empty boats and towers appears even more hollow.
Since the infrastructure is too heavy and large, why not lighten it? Agents born from GPT wrappers are efficient and iterate rapidly in startup and user engagement. Lightweight agents have ample potential to create bubbles, and when the bubbles burst, fertile ground for new life will emerge.
Furthermore, in the current market environment, starting projects with agents and Memecoins can allow products to land in a very short time. Users can directly gain usage experiences, during which agents can cleverly leverage Memecoins to grow the community roadmap, enabling rapid product iteration that is low-cost and fast. Serious AI protocols no longer need to be bound by heavy old consensus frameworks; breaking free from constraints, they can bombard users with lightweight and high-speed iterations. After sufficient market education and dissemination, further enhancements can be made to build foundational infrastructure for grand visions. Lightweight agents cover the ambiguous veil of Memecoins, and community culture and fundamentals will no longer become contradictions, revealing a new asset development path that may emerge as a choice for future AI protocols.
The above discussion addresses the potential for AI agents to become the core narrative. Given the premise that AI agents can continue to grow rapidly, choosing the right ecosystem becomes particularly important. Is it Base? Or Solana? Before answering this question, it might be worthwhile to take a look at the current state of serious agent protocols in the market.
First, Arweave/AO: PermaDAO mentioned that AO adopts the Actor model for design, where each component is an independent agent capable of parallel computation, which highly aligns with the application architecture driven by AI Agents. AI relies on three elements: models, algorithms, and computing power, and AO can meet such high resource demands. AO can independently allocate computing resources for each agent process, effectively eliminating computational performance bottlenecks.
In addition, Spectral is one of the few protocols based on agents, with document-to-code and model inference as its development direction.
Looking back at a type of agent token currently in the market, it can be observed that these agents hardly utilize the underlying infrastructure of the chain. This is a fact because all models in the industry, including agents, operate off-chain. Feeding data is done off-chain, model training is not decentralized, and the output information is not on-chain. This is an objective fact because EVM chains do not support the combination of AI and smart contracts, and of course, neither do Base and Solana. We can expect the introduction of AO next year to see if it can bring models on-chain and perform well. If AO fails, the possibility of models being on-chain may have to wait many years after Ethereum's development, at least not before 2030, or for other public chains to achieve model on-chain capabilities. However, if an architecture like AO with historical resource reserves cannot achieve it, then it may be even more difficult for other public chains.
Currently, AI agent tokens have not many practical use cases. In fact, it is hard to clearly distinguish the difference between AI agent coins and AI Memecoins on Base and Solana. Although agent tokens do not have special uses, why do I believe AI agent coins and AI Memecoins should not be confused? Because I believe we are currently in the stage of creating AI agent bubbles.
Why discuss Base wanting to compete with Solana for the dominant public chain position of AI agents?
In the first half of this bull market, Base has attracted significant market attention. In the competition for market share of Memecoins, Base has had a brief shining performance, such as $BRETT and $DEGEN. However, it still lost to Solana. I believe that AI agents are the next direction for Base to compete in and currently have many advantages.
AI agents will accelerate the birth of bubbles, create chaos, but will ultimately leave users and applications behind:
The birth and expansion of bubbles will attract market attention, and this attention undergoes qualitative changes over time. What are the characteristics of such qualitative changes? In the process of increasing market attention, a series of user pain points and market gaps will be exposed. When the main contradictions cannot be coordinated, but attention continues to increase, that is when qualitative changes are born. When the qualitative change is completed, the sedimented users and applications can undertake grand visions. This is something Memecoins cannot and do not intend to achieve, which is why I believe that although agents and Memecoins are currently ambiguous, they should never be confused.
Before a qualitative change occurs, bubbles will create chaos and various dramas, for example: the number of agents will increase exponentially, and thousands of agents will crowd into users' sight. How to crowd? Agents can integrate with social media like X and Farcaster, self-promoting tokens using various angles that degens love and the unique information density of agents to market the tokens.
Next, fast-iterating agents can complete on-chain transactions, and a group of Viking pirates has invaded the dark forest. Currently, panel protocols in the market, bots in TG groups, and Dune panels will be invaded by agents, where familiar indicators will be manipulated by agents. Trading volume, address numbers, chip distribution, and simulating market maker behavior; on-chain data may require more professional cleaning to reflect value, otherwise, it will be deceived by agents, just like Viking pirates plundering your wealth.
If the market can reach this stage, then the new era belonging to AI agents will have succeeded halfway because "attention equals value" will allow agents to enter the room. This potential comes from:
Strong distribution capability: Agents generate enough topics, such as Goat, and stable distribution paths can be replicated.
Ease of deployment: The deployment platforms for agents will also experience explosive growth. Zerebro, vvaifu, Dolion, Griffain, and Virtual; users can build agents without needing to know any code, and the UX of agent deployment platforms will also optimize in competition.
Memecoin effect: In the startup phase, agent tokens lack suitable business models and have minimal token use cases. Wearing the veil of Memecoins allows for rapid community accumulation, maintaining a high success rate in startups.
The upper limit is extremely high: OpenAI's Level 3 agents are still under development, and products that giants cannot quickly launch will inevitably have vast market space. The lower limit for agents is Memecoins, but the upper limit is autonomous advanced intelligent agents.
Market resistance is low: Agents led by Goat can establish a large audience. Unlike AI infrastructure, users do not feel aversion; when users are not averse, there is ample possibility to start paying attention to it.
Potential incentives: The use cases for agent tokens have not yet been developed. If agents introduce a points system and strengthen incentive measures, they will have the capability to accumulate a large number of users.
Iteration Potential: As mentioned earlier, agents are lightweight and capable of rapidly iterating products. This objective ability to iterate can create increasingly appealing products and content for users.
Therefore, AI agents can become the core narrative and are a battleground for contention.
Why does Base have the potential to compete with Solana?
With the strong support of Coinbase and North American capital, the Base ecosystem experienced explosive growth in 2024. In November, capital inflows exceeded Solana and significantly surpassed Solana in the past 7 days.
If ETH can continue to break through the ETH/BTC exchange rate next year, the spillover effect of the ETH season will have a significant impact on Base. Currently, 23% of the outflow of ETH funds is directed to Base, and this figure is still on the rise.
AI Agent Launchpad Mapping
Virtual
The V1 stage mainly focuses on model training, data contribution, and interaction functions, while in the V2 stage, Virtual launched an AI agent token incubation platform, marked by the release of fun.virtuals in October.
Among them, LUNA has developed into an "independent entity" with its own identity and financial capability. In this process, the roadmap of LUNA aligns with that of Coinbase, which provides powerful technical tools and support to assist in the landing of AI agents on Base.
AI agent technology excels in brand building, especially in creating cultural brands. Through AI agents, brands can interact with communities more efficiently, which includes simplifying interaction tasks and flexibly distributing rewards to enhance user stickiness and brand recognition.
It is worth noting that all transactions of AI agents only support the use of the native Virtual token. The Virtual token absorbs the value capture of the entire ecosystem and becomes an important pillar for ecological development.
Virtual focuses on refining product functionalities, utilizing AI tools to empower users and bridging the gap between Web2 and Web3. It emphasizes 'use value' rather than 'hype.' Although its tool-based products are frequently called upon in practical applications, they lack the dissemination effects typically associated with cryptocurrencies, which is also a shortcoming of the V1 stage.
Clanker
"Posting is Minting" lowers the barrier for token issuance while attracting a large number of users to try. People rush to @Clanker, a phenomenon similar to the operation of summarizing video content by AI on social media; but here, content publication directly translates into asset issuance.
How does Clanker work?
TokenBot (i.e., Clanker) will deploy Meme tokens on Base to a unilateral liquidity pool (LP), with liquidity then locked. Token issuers will gain the following benefits:
0.25% of all swap fees.
1% of the total supply of tokens (with a one-month unlock period).
Users can check the number of tokens deployed or create their own tokens through the official website of clanker.world.
Unlike PumpFun, which issues tokens on Raydium through bonding curves, charging a 1% transaction fee and a fixed fee of 2 SOL; Clanker does not adopt the bonding curve model but collects a 1% fee through Uni v3 trades as revenue.
AI Agent Layer
The AI Agent Layer is a platform within the Base ecosystem focused on creating AI agents and Launchpad, officially launched on November 18. Before the platform's launch, the AIFUN Token was first issued on November 14 and is currently listed on exchanges such as MEXC and Gate, with a current price of $0.09 and a market cap of approximately $25 million.
Creator.bid
Creator.bid was initially an AI platform focused on the monetization and ownership of digital content. In April of this year, the platform completed a new round of financing.
On October 21, Creator.bid announced its official launch on the Base mainnet, enabling one-click creation and publishing of AI agents, providing content creators with new tools and profit models.
Simulacrum
Simulacrum is built on Empyreal. It transforms platforms like Twitter, Farcaster, Reddit, and TikTok into blockchain interaction layers. Users can execute on-chain operations, such as token trading or tipping, through simple social media posts.
Utilizing technologies such as account abstraction, AI agents, intent-driven approaches, and language models to simplify complex blockchain backend operations, making DeFi more accessible to ordinary users.
vvaifu.fun
Similar to Pump.fun, users can easily create AI agents and their associated tokens. AI agents can seamlessly integrate with social platforms like Twitter, Telegram, and Discord, enabling automated user interactions.
Dasha is an AI agent created by vvaifu.fun, with independent Twitter accounts, Telegram channels, and Discord communities, all operated and managed by AI.
Top Hat
Top Hat can not only interact with users through text but also understand and process image content. After a user sends an image, the AI agent can "understand" the image content and respond.
Griffain
With a trainable AI agent platform, Griffain has launched 1,000 trainable AI agents, showcasing the future potential of smart contracts and automated trading.
