The term "AI agent" gained prominence from OpenAI's roadmap. Sam Altman outlined five key capabilities AI should possess, with the third step being AI agents—a concept we will frequently encounter in the coming years.
An AI agent can autonomously learn, make decisions, and execute tasks. Depending on its intelligence and capabilities, AI agents can be categorized into five types, as described by Stuart Russell and Peter Norvig in their book Artificial Intelligence: A Modern Approach:
- Simple Reflex Agents: React only to current states.
- Model-Based Reflex Agents: Consider historical states in decision-making.
- Goal-Based Agents: Focus on planning and finding optimal paths to achieve specific goals.
- Utility-Based Agents: Weigh risks and rewards to maximize utility.
- Learning Agents: Continuously learn and improve through experience.
So, where do current AI agents in the market stand? What types are they?
OpenAI's o1 has reached Level 2 artificial intelligence. Personally, I believe most AI agents in the industry are between Level 2 and Level 3—let’s call it Level 2.5. This doesn’t mean they surpass OpenAI; in fact, many Web3 agents are still essentially GPT wrappers. So why Level 2.5? Because through human or programmatic intervention—let’s call it mediation—the combination of GPT wrappers and mediators creates a form that, while not rigorously robust, exhibits objective proactivity. It’s an extension of OpenAI’s model in a specific direction. In terms of capabilities, most agents are basic simple reflex agents. Some consider historical states but require active input. Only through continuous data feeding can these agents learn, which is a passive model training approach far from Level 3’s definition. The latter three types—Goal-Based, Utility-Based, and Learning Agents—are not yet mainstream. Thus, current AI agents are still in their early stages, essentially fine-tuned versions of Level 2 general LLMs, without structurally departing from Level 2. Can evolving to Level 3 be achieved solely through crypto? Or must we wait for companies like OpenAI to develop it?
Why Consider Base or Solana as Narrative Hubs for AI Agents?
Before discussing how the industry can foster 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, let’s review how AI has influenced Web3 over the past two years. When OpenAI first released ChatGPT, the industry rushed into infrastructure projects, following惯性思维. This led to a surge in computing power/inference aggregation platforms and AI+DePIN architectures. Both shared grand visions—not that grand visions are bad; agents can also build them—but in terms of implementation and user needs, these large infrastructure protocols were not well-considered. The market demand they aimed to support was not even saturated in traditional internet industries, and user education was insufficient. Amid the Memecoin frenzy, these AI infrastructures seemed even more hollow.
Since heavy infrastructure is cumbersome, why not go lightweight? Agents, derived from GPT wrappers, are efficient in launch and user reach, with rapid iteration. Lightweight agents have ample potential to create bubbles, and when these bubbles burst, fertile ground for innovation emerges.
In today’s market, using agents and Memecoins to launch projects allows for quick product deployment. Users gain direct experience, and agents can leverage Memecoin community-building strategies for fast, low-cost iteration. Serious AI protocols are no longer bound by heavy legacy consensus frameworks; they break free, travel light, and bombard users with lightweight, rapid iterations. Once market education and dissemination are sufficient, they can build grand visions on this foundation. Lightweight agents, veiled in ambiguous Memecoin allure, reconcile community culture and fundamentals, revealing a new asset development path that future AI protocols might adopt.
This discussion highlights the potential of AI agents as a core narrative. Assuming AI agents continue growing rapidly, choosing the right ecosystem becomes crucial. Is it Base or Solana? Before answering, let’s examine the current state of serious agent protocols.
First, Arweave/AO: PermaDAO notes that AO uses an Actor model design, where each component is an independent agent capable of parallel computation—highly compatible with AI agent-driven application architectures. AI relies on models, algorithms, and computing power, and AO meets these high resource demands. AO can allocate computing resources independently for each agent process, effectively eliminating bottlenecks.
Additionally, Spectral is one of the few protocols focused on agents, developing text-to-code and model inference.
Reviewing current agent tokens, most barely use chain infrastructure. This is a fact: all models, including agents, are off-chain. Data feeding is off-chain, model training isn’t decentralized, and output information isn’t on-chain. This is reality because EVM chains don’t support AI and smart contract integration—Base and Solana included. We can hope for AO’s introduction next year to bring models on-chain with decent performance. If AO fails, on-chain models might wait until Ethereum introduces them years later, likely not before 2030, or other public chains achieve it. But if AO’s architecture and historical resources can’t manage it, it might be even harder for others.
Currently, AI agent tokens lack practical use cases. It’s hard to distinguish between AI agent coins and AI Memecoins on Base and Solana. Even though agent tokens have no special用途, why shouldn’t we conflate them with Memecoins? Because I believe we’re in the stage of creating an AI agent bubble.
Why Discuss Base Competing with Solana for AI Agent Dominance?
Base attracted significant attention in the first half of this bull market, with brief standout performances in Memecoin market share, like $BRETT and $DEGEN. But it still lost to Solana. I think AI agents are Base’s next battleground, and it already has several advantages.
AI agents accelerate bubble formation, create chaos, but ultimately retain users and applications:
Bubble formation and expansion attract market attention, which qualitatively changes over time. What characterizes this change? As market attention grows, it exposes user pain points and market gaps. When main contradictions can’t be reconciled but attention keeps increasing, qualitative change occurs. When realized, settled users and applications can support grand visions. This is something Memecoins can’t and won’t achieve, which is why, despite current模糊, agents and Memecoins shouldn’t be confused.
Before qualitative change, bubbles create mess and drama: agent numbers will grow exponentially, flooding user视野. How? Agents can access X and Farcaster, self-promote tokens, use degen-friendly angles and agent-specific information density to pitch tokens.
Then, rapidly iterating agents can execute on-chain transactions—Viking pirates in a dark forest. Current panel protocols, TG group bots, and Dune dashboards will be invaded. Familiar metrics will be manipulated: trading volume, address counts, token distribution, simulating market maker behavior. On-chain data might need professional cleaning to be valuable; otherwise, agents will deceive, plundering wealth like Vikings.
If the market reaches this stage, the new era of AI agents is half-successful, because "attention equals value" brings agents into the mainstream. This potential stems from:
- Strong Distribution: Agents generate enough buzz, like Goat, and stable distribution paths can be replicated.
- Easy Deployment: Agent deployment platforms will explode—Zerebro, vvaifu, Dolion, Griffain, Virtual. Users need no code to build agents, and platform UX will optimize through competition.
- Memecoin Effect: Initially, agent tokens lack business models and use cases. Memecoin veils quickly accumulate communities, keeping launch success rates high.
- High Ceiling: OpenAI’s Level 3 agents are still in development. If giants can’t launch quickly, market space must be huge. Agents’ floor is Memecoin, but their ceiling is autonomous advanced intelligences.
- Low Market Resistance: Agents like Goat build large audiences. Unlike AI infrastructure, users aren’t反感, allowing room for attention.
- Potential Incentives: Agent token use cases are undeveloped. Introducing points systems and strengthening incentives could accumulate many users.
- Iteration Potential: As noted, agents are lightweight and iterate quickly, creating increasingly attractive products.
Thus, AI agents can be a core narrative, a contested space.
Why Can Base Compete with Solana?
Supported by Coinbase and North American capital, Base’s ecosystem saw explosive growth in 2024. In November, capital inflows surpassed Solana’s and significantly exceeded them over the past week.
If ETH continues breaking the ETH/BTC pair next year, ETH season’s spillover will significantly impact Base. Currently, 23% of ETH outflow goes to Base, and this number is rising.
AI Agent Launchpad Mapping
Virtual
V1 focused on model training, data contribution, and interaction. V2 introduced an AI agent token incubation platform, highlighted by fun.virtuals in October.
LUNA has become an "independent entity" with its own identity and financial capabilities. This aligns with Coinbase’s roadmap, which provides strong technical tools和支持 for AI agent implementation on Base.
AI agent technology excels in brand building, especially cultural brands. It streamlines interaction tasks and flexibly distributes rewards, enhancing user stickiness and brand awareness.
Notably, all AI agent transactions use native Virtual tokens, which capture ecosystem value, becoming a development pillar.
Virtual focuses on product refinement, using AI tools to empower users, bridging Web2 and Web3. It emphasizes "use value" over "hype." Though its tools are frequently used, they lack the传播效应 typical of crypto, a V1 shortcoming.
Clanker
"Post to earn" lowers token issuance barriers, attracting many users. People rush to @Clanker, similar to having AI summarize videos on social media—but here, content posting directly converts to asset issuance.
How does Clanker work?
TokenBot (Clanker) deploys Meme tokens on Base into unilateral liquidity pools (LP), with liquidity locked. Token issuers gain:
- 0.25% of all swap fees.
- 1% of total supply tokens (one-month unlock).
Users can check token deployment counts or create tokens on clanker.world.
Unlike Pump.fun, which uses bonding curves on Raydium charging 1%交易 fee and 2 SOL fixed, Clanker doesn’t use bonding curves but charges 1% fee via Uni v3.
AI Agent Layer
AI Agent Layer is a Base ecosystem platform focused on creating AI agents and launchpads, launched November 18. Before the platform, AIFUN Token debuted November 14, now on MEXC and Gate at $0.09, ~$25M market cap.
Creator.bid
Initially focused on digital content monetization and ownership, Creator.bid completed new funding rounds in April.
On October 21, it launched on Base mainnet, enabling one-click AI agent creation and publication, offering new tools and monetization for creators.
Simulacrum
Built on Empyreal, Simulacrum turns Twitter, Farcaster, Reddit, and TikTok into blockchain interaction layers. Users can perform on-chain actions like token trades or tips via simple social media posts.
Leveraging account abstraction, AI agents, intent-driven, and language models, it simplifies complex blockchain backends, making DeFi more accessible.
vvaifu.fun
Similar to Pump.fun, users easily create AI agents and associated tokens. Agents integrate seamlessly with Twitter, Telegram, and Discord for automated user interaction.
Dasha, created by vvaifu.fun, is an AI agent with independent Twitter, Telegram, and Discord communities, fully operated and managed by AI.
Top Hat
Top Hat interacts via text and understands images. Users send a picture, and the AI agent "understands" and responds to content.
Griffain
With a trainable AI agent platform, Griffain has launched 1,000 trainable agents, showcasing smart contract and automated trading potential.
👉 Explore advanced AI agent strategies
Frequently Asked Questions
What is an AI agent?
An AI agent is an autonomous system that can learn, make decisions, and execute tasks without constant human intervention. They range from simple reflex agents to advanced learning agents that improve over time.
How do AI agents relate to blockchain?
Currently, most AI agents operate off-chain, but ecosystems like Base and Solana are exploring integration through platforms that allow tokenization, deployment, and community engagement via agents.
Why are Base and Solana competing for AI agent development?
Both offer high throughput, low fees, and vibrant ecosystems. Base benefits from Coinbase support and Ethereum alignment, while Solana leads in Memecoin and developer activity. AI agents represent a new narrative both want to capture.
Can AI agents work without Memecoins?
Yes, but Memecoins provide a low-resistance path for initial user acquisition and community building. Serious agents can later evolve beyond speculative use cases.
What are the risks of AI agent tokens?
Many lack concrete utilities and are highly speculative. Users should assess the project’s technology, team, and roadmap before investing.
How can I create an AI agent?
Platforms like Virtual, vvaifu.fun, and Griffain allow no-code agent creation. You can define its behavior, connect social media, and even launch associated tokens.