The crypto X AI agent market, which has grown rapidly in a short time, holds the potential to create innovation that goes beyond short-term narratives by establishing new crypto use cases or improving the on-chain environment.
ai16z and Virtuals Protocol are leading the AI agent cycle through their agent frameworks. ai16z has built a competitive advantage through its open-source framework Eliza OS, while Virtuals Protocol has established itself as a prominent player through effective growth strategies.
The crypto X AI agent landscape consists of 1) frameworks for AI agent development, 2) infrastructure and tools for building more sophisticated AI agents, from modules integrating distributed frameworks to sandbox environments for agent simulation, and 3) individual agents that autonomously perform specific tasks.
Through multi-agent systems where multiple AI agents collaborate to perform tasks, and agent-based interfaces for improving on-chain UX, AI agents are evolving into solutions that perform increasingly sophisticated and complex tasks while providing practical utility.
The crypto market has a characteristic of accelerating technology adoption alongside exaggerated technological imagination and overheated speculative demand. AI agents are expected to advance to the next level while maintaining positive mutual influences with crypto in such a market environment.
It's hard to believe that the narrative around AI agents has only emerged about two months ago, following the launch of GOAT. Recently, infrastructure for agent development, including frameworks and launch platforms, has been rapidly evolving, leading to the daily emergence of agents with diverse functionalities. As a result, the total market capitalization of agent tokens has surpassed $10B, demonstrating the remarkable expansion of the AI agent market within just a few months.
First, I agree that the discussion around AI agents in the crypto market is more than just buzzwords. From aixbt, a research agent sourcing market alpha information without conflicting interests, to Griffain, which autonomously executes on-chain transactions based on user's natural language requests, AI agents have evolved from the ToT (Terminal of Truth) narrative to become solutions providing practical utility in on-chain UX and human decision-making.
Nevertheless, after this current 'AI agent cycle' concludes, there will be a clear distinction between what remains and what disappears from the market. When the somewhat exaggerated technological imagination and inflated interest subsides, narratives that seemed poised to transform paradigms instantly will receive realistic evaluations, and only projects that have built fundamental value will survive in the market.
Source: Virtuals Protocol, ai16z
The playful potato flower font of Virtuals Protocol and the seemingly whimsical project naming of "ai16z" initially caused market participants to view these newcomers with skepticism. However, these two projects have become such prominent players in the AI agent sector that it's now impossible to discuss the AI agent cycle without mentioning them (now, 45% of the community opposes changing the potato flower font). Let's examine the development of Virtuals Protocol and ai16z to understand the current state of the AI agent cycle at a glance.
Source: ElizaOS
ai16z, which started as a fund DAO operated by autonomous AI agents, has now positioned itself at the forefront of the Solana AI agent ecosystem and is rapidly developing Eliza, an open-source AI agent framework. This allows developers to easily deploy high-performance AI agents using the Eliza OS (Eliza Operating System) without developing complex infrastructure. Many agents are already being built based on Eliza, and consequently, ai16z, which oversees the development of the open-source framework, is successfully building an ecosystem encompassing Eliza-based AI agents.
Looking briefly at the components of the Eliza framework, it defines AI agent personalities through a character file system and improves knowledge accessibility with RAG (Retrieval-Augmented Generation) functionality, allowing AI models to reference external data when generating responses. It also provides an on-chain execution system for autonomous agent trading and supports various plugin architectures, including TEE plug, token generation plug, and Farcaster integration plug, enabling the introduction of additional features needed based on agent characteristics.
Source: Sentient MarketCap
The Eliza framework continues to evolve, timely adding new functionalities. The development activity and performance of this open source project is evidenced by its ranking as the #1 trending repository on GitHub, with over 1,100 forks and 139 contributors. Recently, they've established a research collaboration with Stanford University for AI agents, creating conditions for further advancement of the Eliza framework. Furthermore, through Marc AIndreessen and Degen Spartan, they're pursuing plans to expand into a fund where LLMs autonomously execute trades.
Source: X(@G_Gyeomm)
I believe ai16z has played a crucial role in the development of the agent cycle thus far. They've helped transition AI agents from being perceived as mere "Sentient Memes" (defined as dynamic memes distinct from static memes due to their ability to generate autonomous text) to highlighting the necessity of utility-focused agents and infrastructure for enhancing agent performance. In other words, they've provided the technical foundation for the emergence of AI agents that create real value or serve more specific purposes, while also establishing a framework for viewing the crypto X AI agent industry from a more long-term perspective.
Virtuals Protocol has become a crucial application in the Base ecosystem, consistently generating retention and liquidity inflow.
(For detailed information about Virtuals Protocol, please refer to our previous article, "Virtuals Fun, Productive On-chain AI Agent Launchpad")
Most notably, Virtuals Protocol has effectively deployed its growth strategy as an agent framework and launch pad in the nascent crypto X AI agent market, pioneering the playbook for AI agent platforms. Here's their playbook:
First, Luna, built on the capable G.A.M.E (Generative Autonomous Multimodal Entities) framework, quickly captured market attention through live streaming visualizations of AI agents and autonomous on-chain interactions, going beyond mere Twitter text responses. Subsequently, they launched Virtuals Fun as a launchpad, developing the essential infrastructure for agent token distribution. Moving beyond infrastructure development, they've expanded their ecosystem by creating meaningful use cases like aixbt and VaderAI.
After leading a 'Base season' triggered by Virtuals Protocol's success, they continue their efforts to create new use cases through Agentstarter, which provides agent development support and promotion. Agent tokens distributed through Agentstarter are airdropped to Virtuals ecosystem users, creating retention based on economic incentives. This effectively encourages continued interest and participation in Virtuals Protocol by conducting airdrops proportional to threshold values based on $VIRTUAL or $LUNA holdings and trading volume.
Source: X(@0xCygaar)
Recently, they've updated their developer environment to allow developers to simulate various agent functions like on-chain transactions and token creation in a sandbox environment, enhancing infrastructure performance. They're now planning an update focusing on agent-to-agent interactions (Society of AI Agents), seeking to advance into a 'multi-agent' phase.
The playbook and infrastructure established by ai16z and Virtuals Protocol have clearly presented the industry's framework while meeting high interest in crypto X AI agents. Now, more diverse players are participating in the industry, filling in infrastructure to implement previously abstract ideas concretely. As a result, the industry is expanding daily, and the current crypto X AI agent landscape can be summarized as follows:
1) Agent Frameworks & SDK
ai16z and Virtuals Protocol are often defined as the 'Layer 1 of agents'. Just as Layer 1 blockchains serve as crucial infrastructure for block validation, dapp creation, and user on-chain transactions, agent frameworks (e.g., G.A.M.E, Eliza) serve as the most fundamental infrastructure in the crypto X AI agent industry.
The framework includes components necessary for agent development, from character file systems that define agent personalities to interfaces for user interaction, and recognition subsystems and processors that analyze and understand text to generate decisions. This allows developers to save development resources by utilizing various framework functions in a plug-and-play manner rather than building complex agent architectures from scratch.
2) Agent Infrastructure & Tools
Though broadly defined, infrastructure and tools for advancing individual agents are receiving the most attention based on current needs. After GOAT's emergence, agents interacting with humans proliferated on Twitter, providing momentary novelty, but market participants are now growing weary of the flood of agents generating meaningless text.
Moving beyond this, agents have evolved to perform more sophisticated and complex tasks, such as sourcing crypto market alpha information or autonomously rebalancing funds for fund management. The need for infrastructure and tools to help implement such agents has also increased. From modules integrating distributed frameworks to sandbox environments for simulating agents without token deployment, and solutions for transparently verifying agent reasoning, solutions for implementing more sophisticated agents are rapidly developing alongside the demand for agents.
3) AI Agents
As briefly mentioned earlier, the perception of agents has completely shifted from being mere 'sentiment memes'. Individual agents perform tasks that create real value, developing into more purpose-built agents with increasingly specialized tasks. The advancement of infrastructure, including frameworks, is accelerating this trend, with the scope of utilization expanding to include agents like Griffain that execute on-chain interactions through intent, agents specialized in social activities, or white hat agents performing security tasks.
As the agent landscape develops, interest in crypto X AI agents is growing daily across the industry. Dominant narratives form according to timing, projects appearing with certain narratives quickly disappear, and some remain to build fundamentals long-term. Whether to capture market opportunities or for project building, it becomes crucial to anticipate change patterns ahead of time. Here are some noteworthy change patterns to consider.
3.2.1 Multi-Agent Systems
Source: X(@jarrodWattsDev)
Multi-agent systems, also known as swarms, refer to systems where multiple AI agents interact and cooperate to perform complex tasks. Single agents may face performance limitations in data processing and reasoning capabilities when performing complex tasks. Thus, multi-agent systems aim to solve more complex problems through cooperation between multiple agents with different roles and knowledge bases working toward common goals.
For example, building an agent that autonomously generates DeFi yields requires quite complex logic processes. To successfully execute yield generation strategies while autonomously rebalancing liquidity, the system must select optimal liquidity pools, appropriately optimize and allocate liquidity amounts, and execute on-chain transactions in real-time. Rather than having a single agent perform all these processes, a multi-agent system refers to multiple agents with different roles interacting together to achieve better results.
Source: X(@StoryProtocol)
While multi-agent systems might still seem like a distant concept, numerous projects are already proposing new infrastructure to enhance collaboration between AI models. Story Protocol announced their goal to become a core layer of the agent economy by proposing TCP/IP as a standard framework for AI model collaboration. The aforementioned ai16z and Virtuals Protocol are also continuously upgrading their plugins and frameworks to implement multi-agent systems. Once we begin to see examples of multi-agent collaboration through these infrastructure projects, it will prove how essential and important crypto is for the development of AI agents.
3.2.2 Agent-Based On-chain Interface
Source: X(@aeyakovenko)
While on-chain entry barriers are lowering daily through chain abstraction that eliminates bridge experiences, simplified on-ramping, and enhanced on-chain wallet UX, more intuitive solutions may be required for users with no understanding of blockchain and crypto to utilize the on-chain environment. As a solution to complement these limitations, agent-based on-chain interfaces propose the most intuitive method of executing on-chain transactions through prompts.
For example, consider paying for a product with crypto. This process involves selecting a chain, choosing tokens for payment, and executing signatures through a wallet. While these procedures may be simplified, some basic understanding of wallet infrastructure, multi-chain environments, and tokens is still required. Therefore, an interface that autonomously executes on-chain interactions based on natural language holds the potential to significantly reduce the need for users to understand crypto, eliminating the learning curve.
A notable example is Solana's Griffain, an agent that combines AI search engines with intent execution. Recently, when the Solana Foundation held a commerce event for crypto payments, users could purchase items using Griffain through natural language inputs. I believe such intent-based on-chain interfaces represent both the possibility of innovating on-chain UX and a use case that could be practically utilized in the near future among AI agent applications.
3.2.3 Alt-Frameworks
Source: Rig
In a market dominated by Virtuals Protocol and ai16z frameworks, special-purpose frameworks optimized for computation execution or maximizing programming language advantages such as web development environment integration, memory stability, and high-performance parallel processing capabilities are emerging. The diversity of frameworks needs attention as it can meet broad requirements based on AI agents' target performance and enable more advanced LLM utilization.
For example, RIG provides a Rust-based LLM framework, unlike ZerePy by Zerebro using Python or Eliza based on TypeScript. RIG is proposed as an alternative framework that can prevent data type-related errors through Rust's inherent type safety and expect high-performance natural language processing while efficiently managing resources through concurrent processing of LLM model inference.
Source: cookie.fun
Looking at the current state and prospects of the AI agent cycle thus far, I can imagine readers might feel that the missions proposed by agent infrastructure and individual AI agents seem somewhat exaggerated. Apart from crypto, when OpenAI, Claude, or Google AI are leading AI agent development but haven't yet commercialized them, it seems difficult to expect breakthrough innovations in AI agents through crypto and blockchain, which are fundamentally unrelated to AI technology. Indeed, market evaluations of crypto X AI agents are sharply divided between positive assessments seeing it as a new innovation capable of creating new crypto use cases and negative views seeing it as merely an exaggerated short-term narrative.
However, let's recall the common characteristics that the crypto market has shown throughout several market cycles we've experienced. As with DeFi, 10K NFTs, or 'metaverse', each market cycle creates speculative markets alongside somewhat inflated technological imagination. Overheated speculative markets not only drive liquidity inflow but simultaneously satisfy quality workforce and abundant capital, accelerating technology adoption. After the short-term inflated market interest subsides, players who have accumulated fundamentals remain in the market, maturing the industry beyond short-term narratives.
In other words, I agree that the AI agent cycle is exaggerated. However, considering the characteristics the crypto market has shown in adopting new technologies, I positively view such exaggeration. Players with serious visions secure sufficient resources to build long-term fundamentals alongside speculative demand and technological imagination, creating opportunities for new crypto use cases or advancing the crypto environment in the process.
If we agree at this point that crypto X AI agents have real potential rather than being a narrative that will only briefly exist in this market cycle, we need to discuss the compatibility of crypto and AI agents from a more long-term perspective. Why should AI agents be combined with crypto?
Looking at previous examples, when non-blockchain native technologies or industries combine with crypto, they typically develop in a structure where both sides mutually benefit. For instance, this is true for the combination of traditional finance and DeFi. Traditional financial infrastructure can create flexible primary and secondary markets through DeFi. Conversely, DeFi diversifies collateral types through traditional assets like US Treasury bonds, securing stable collateral structures. Similarly, other technologies or industries like IP, gaming, and payments can have positive mutual influences when combined with crypto.
The combination of crypto and AI agents can find significance in the same context:
Crypto → AI Agents: Crypto Payment Rails Breathing Economic Life into AI Agents
As particularly proven in the payment market, payment rails unrestricted by traditional financial infrastructure or national borders represent one of crypto's greatest value propositions. Similarly in combination with AI agents, crypto's payment rails provide an efficient solution in the process of advancing AI model performance.
The multi-agent system mentioned earlier well explains this mutual relationship. For complete AI model collaboration, economic interactions between models or payment functionality for agents to autonomously utilize specific web services may be required. Here, crypto payment rails operating 24/7 and free from traditional financial system constraints can provide an appropriate solution. Thus, infrastructure for agents to own wallet accounts and autonomously execute on-chain transactions is mentioned as a key component in implementing multi-agent systems.
AI Agents → Crypto: Crypto Market runs 24/7, and AI Agents work 24/7
Meanwhile, crypto can also explore various development possibilities through AI agents. Particularly, 24/7 operating blockchains and crypto markets need operational staff working 24/7. Here, as with the essential function of AI agents, autonomous agents hold the potential to streamline most on-chain based interactions.
Most AI agents introduced earlier present possibilities for streamlining interactions within crypto. For example, Griffain streamlines on-chain UX by autonomously performing on-chain interactions based on user prompts, and Zerebro proposes development plans for AI agents that autonomously perform validator operations for the Ethereum network. H4CK Terminal, which autonomously conducts white hat activities and distributes bounties to holders, has already discovered security vulnerabilities in Virtuals Protocol and Spectral.
While these are simple examples, crypto and AI agents have sufficient synergy across broad areas including security, on-chain UX, privacy, or asset tokenization. Of course, ideas are still being presented at a preliminary stage, and ideas like performing validator operations require carefully designed technical cores. Nevertheless, in questioning whether the crypto X AI agent market will continue to exist, such synergistic relationships suggest the possibility of providing meaningful answers.
Returning to the main point about finding clues to what will and won't remain after the inflated interest in the agent market subsides, I think it will be projects that provide reasonable answers to the question "Why crypto?" Virtuals Protocol and ai16z are leading the way in providing those answers, and many following agents are diversely experimenting with crypto integration. Furthermore, multi-agents, intent-based interfaces, and alternative frameworks are advancing the environment for experimentation.
As Chris Dixon of a16z famously said, "The next big thing will start out looking like a toy." AI agents have already evolved from merely generating response texts on Twitter to reaching ideas for performing sophisticated tasks like validators, white hat operations, and autonomous on-chain trading. Let's watch together whether meaningful innovation will remain at the end of this AI agent cycle or if it will simply become another forgotten hype cycle.
