The Second-Half of AI Agent: The Force Awakens

The cryptocurrency market is experiencing renewed momentum in May 2025, marked by Bitcoin's surge to an all-time high of over $110,000, driven by increased institutional investment and regulatory clarity. Amid this bullish backdrop, we think the force of the AI Agent sector is awakening, driven by rapid advancements in artificial intelligence and the growing integration of AI with blockchain technology. In this artcicle, CoinEx Research will attempt to examine the burgeoning developments within this space, focusing on key infrastructure initiatives such as the Virtuals Protocol and the ai16z-backed ElizaOS ecosystem. We will also analyze their respective strategies for fostering AI Agent ecosystems on Base and Solana, scrutinize the underlying mechanisms driving their growth, and explore broader potential directions for blockchain projects capitalizing on the AI narrative, with a particular emphasis on the pivotal role of Model Context Protocol (MCP) infrastructure in enabling sophisticated AI agent functionalities within the decentralized landscape.

You can also review our previous research: AI Agents in the Crypto World: Revolutionary Evolution from Web2 to Web3

Virtuals Protocol - Ecosystem Follow-up

The Virtuals Protocol has maintained a consistent trajectory of development, strategically building the infrastructure for a sophisticated AI Agent ecosystem on the Base network. Key milestones achieved in recent months underscore this commitment to a phased rollout:

• February 2025: Agent Commerce Protocol (ACP) launch
• March 2025: Virtuals Partner Network (VPN) rollout
• April 2025: Genesis Launch platform debut

Agent Commerce Protocol (ACP) Deployment

The introduction of the ACP represents a foundational layer for autonomous AI Agent interaction. This multi-agent framework establishes a standardized environment for secure, verifiable, and efficient exchange and collaboration through a well-defined operational cycle:

• Request Phase: Agents establish initial contact request and determine basic compatibility for a transaction
• Negotiation Phase: Agents agree on specific terms, which are cryptographically signed to create a Proof of Agreement (PoA)
• Transaction Phase: The actual exchange of value occurs, with both payment and deliverables held in escrow
• Evaluation Phase: The transaction is assessed against the agreed terms, enabling reputation building and continuous improvement

Source: Virtuals Protocol

The ACP effectively constructs a decentralized commercial ecosystem for AI Agents, enabling autonomous interaction, collaborative endeavors, and seamless transactional capabilities. The Virtuals Protocol illustrates the application of ACP through a simulated study involving five independent, specialized agents collaborating to establish and operate a toy lemonade stand business. This interactive multi-agent demo offers users tangible insights into the autonomous coordination facilitated by the ACP. 

In fact, Google's subsequent announcement of a conceptually similar Agent-to-Agent (A2A) framework, while differing in its reliance on protocol-level connections rather than direct smart contract interaction, underscores the growing industry recognition of this paradigm.

Virtuals Partner Network (VPN) Establishment

The VPN strategically aggregates a diverse cohort of stakeholders, including investors, pioneering founders, domain experts, and academic researchers, with the primary objective of sourcing, supporting, and scaling high-potential AI x Crypto ventures. This network fosters a synergistic ecosystem where the collective expertise and capital of its members contribute to the accelerated growth and maturation of promising projects, effectively functioning as an incubation hub for innovative builders and project teams.

Genesis Launch Introduction

The Genesis Launch represents an innovative paradigm for the initial distribution of AI Agent tokens, drawing conceptual parallels with platforms like pump.fun but with a significant pivot towards contribution-based participation. This launchpad signifies an evolution in the "initial DEX offering" (IDO) model. In contrast to the often adversarial and bot-dominated environment of platforms like pump.fun, Genesis Launch prioritizes demonstrable contributions to the ecosystem. Users accumulate points through various engagement mechanisms to gain preferential access to pre-sales of novel AI Agent tokens, thereby incentivizing active participation and value creation within the Virtuals ecosystem.

Source: Virtuals Protocol

Current mechanisms for accruing points include:

• Investing in AI Agent Projects: This primarily involves strategic investment in existing AI Agents within the Virtuals ecosystem, currently categorized as Sentient and Prototype. This mechanism effectively provides a form of point accrual proportional to investment, which can then be redeployed for participation in Genesis Launch events.
• Holding $VIRTUAL: Direct holding of the native $VIRTUAL token accrues points, with potential future enhancements through staking mechanisms.
• Agent Staking: Staking the native token of recognized agents within the Virtuals ecosystem, provides additional point accrual, incentivizing support for a variety of projects.
• "Yapping Points": Recognizing the value of community-driven promotion, users can earn points by actively engaging and creating content related to Virtuals on Twitter, a mechanism familiar to users of Kaito AI.

The tokenomics structure of Genesis Launch incorporates a transparent and anti-monopolistic 24-hour pre-sale window:

• Allocation: 37.5% of the total new token supply is allocated for the pre-sale, 12.5% is designated for liquidity pool injection, and the remaining 50% is allocated for project development, treasury management, and marketing initiatives.
• Dynamic Distribution: Individual allocation within the pre-sale is dynamically determined by the proportion of a participant's accumulated points relative to the total points pool, with a capped individual allocation of 0.5% of the total supply to mitigate the risk of whale accumulation. The system calculates individual allocations in real-time based on the total points contributed by all participants.
• Refund Mechanism: Any unutilized $VIRTUAL tokens and accrued points are automatically refunded to participants.
• Fixed Valuation Launch: New projects initiated through Genesis Launch adopt a fixed initial market capitalization and total supply model, standardized at a base valuation of 112,000 $VIRTUAL tokens (approximately $200,000 fully diluted valuation).

Source: Virtuals Protocol

Given the relative stability of Virtuals' underlying technology and core narrative, this novel 'new-for-old' asset issuance model has effectively sparked interest in a subdued market. By incentivizing participation through a contribution-based system, this approach revitalizes liquidity and engagement within the Virtuals ecosystem. Zooming out to the broader market, considering the broader market context, it is observable that previously prominent AI Agent tokens have experienced a degree of upward price correction.

Virtuals Protocol Ecosystem Spotlight: Basis OS

A notable success story originating from the Genesis Launch is Basis OS. Functioning with similarities to Ethena in its yield-generating approach, Basis OS is a decentralized protocol engineered to optimize returns in specialized markets through a managed basis trading strategy. The core mechanism involves simultaneously establishing a long position in the spot market and a short position on perpetual decentralized exchanges, enabling the protocol to capture funding rate premiums while maintaining a market-neutral exposure.

Virtuals Protocol Ecosystem Spotlight: Arbus Terminal

Arbus is another potential project launched via Genesis Launch, delivering an AI-powered market intelligence layer for the InfoFi and agentic economy. The Arbus Terminal, its flagship platform, translates fragmented Web3 signals into structured, real-time insights, empowering traders, developers, and AI agents with actionable data. By integrating with Virtuals’ Agent Commerce Protocol (ACP), Arbus enables autonomous agents to coordinate and transact on-chain with enhanced contextual intelligence, fostering smarter decentralized applications and governance.

Source: Arbus

ElizaOS / ai16z - Ecosystem Follow-up

ElizaOS V2 Development

The ai16z team introduced ElizaOS version 2, designed to provide a leaner, more flexible architecture for cross-platform agent interactions. The major improvements include: 

• Cross-Platform Presence: Your agent can jump from Twitter, continue via SMS, and place orders with businesses—all while maintaining conversation context.
• Unified Blockchain Management: One wallet to rule all chains—no more chain-switching headaches.
• Autonomous Workflows: Agents that handle multi-step processes independently—researching data, analyzing results, generating reports, and scheduling follow-ups without constant oversight.
• Evolving Intelligence: Database-driven characters that can start minimal and grow through interactions with your crowd.
• Enhanced Security: Native TEE integration provides verifiable security guarantees for sensitive operations while maintaining privacy

auto.fun Platform Debut

Similarly, the ai16z team has introduced auto.fun, a platform for launching AI agents and their associated tokens on the Solana blockchain. This platform aims to foster a more equitable and sustainable model for agent tokenization, incorporating features such as customizable contract addresses, AI-verified communication channels, and community content creation tools.

Key features of auto.fun include:

• "Fairer than fair launch": Implementing a bonding curve mechanism allows project teams to secure up to 50% of their token supply prior to public market listing. This approach aims to address the inherent limitations of pure fair launches while ensuring broader community access rather than privileged insider allocations.
• No-code agent builder: Integration with Fleek enables creators to deploy agents alongside their tokens with minimal or no coding requirements or to seamlessly connect existing agents to their corresponding tokens.
• AI-generated marketing: The platform facilitates the automated generation of tokens, marketing content, and visual assets based on a single user prompt, streamlining the launch process.
• Sustainable project funding: A Liquidity NFT mechanism ensures that projects accrue ongoing revenue from trading fees generated on decentralized exchanges, reducing their reliance on initial token sales for long-term development funding.

Tokenomics Structure of auto.fun:

• Current Mechanics - Primary SOL:AT Pool: Upon an Agent Token (AT) graduating from the auto.fun platform to Raydium, the associated liquidity is algorithmically burned, generating a non-fungible token representing this burned liquidity (Liquidity NFT). This Liquidity NFT is subsequently paired with SOL to establish the primary liquidity pool (SOL:AT) on Raydium.
• Liquidity NFTs for Project Creators (Powered by Raydium): Each project successfully launching on auto.fun is issued a Raydium Liquidity NFT, which represents ownership of the future trading fees generated from the burned liquidity on Raydium.
• Platform Fees → $ai16z Buybacks: A designated portion (10%) of the burned LP, denominated in SOL, is allocated upon token graduation. The fees generated from this SOL are then utilized to conduct buybacks of the platform's native $ai16z token from the open market.

Source: ElizaOS

• Future Phase (Under Development) - Secondary $ai16z:AT Pools: The platform's architectural roadmap includes the introduction of a secondary layer of liquidity through $ai16z:AT pools. These pools are currently under development and are intended to deepen liquidity for both $ai16z and the successful ATs over their lifetime.

Despite the introduction of above features, we think the core token issuance functionality offered by auto.fun appears to largely emulate the established operational models of existing platforms such as pump.fun. The token creation process, issuance mechanisms, and overall user experience do not currently exhibit significant differentiating characteristics. Furthermore, the market response to tokens launched on auto.fun has been generally muted, failing to inject substantial incremental value into the broader ai16z ecosystem. The high degree of content and functional similarity has also elicited skepticism and negative feedback from segments of the community. 

Potential Directions for Blockchain Projects under the AI Narrative: Focusing on MCP infrastructure and applications

The performance and utility of Large Language Model (LLM) applications are fundamentally governed by three critical factors: the inherent capabilities of the LLM model itself, the quality and relevance of the data it is trained upon, and the precision and context of the prompts used to elicit desired outputs. Among these, data is particularly crucial. Standardized LLMs are susceptible to the "garbage in, garbage out" phenomenon, underscoring the imperative of high-quality, contextually relevant data for generating meaningful and accurate outputs.

Within the cryptocurrency domain, the development of highly specialized LLMs tailored to the nuanced intricacies of blockchain data represents a significant undertaking in terms of computational resources and development costs, potentially limiting its widespread feasibility for many projects. Consequently, innovation at the data layer, specifically through the implementation of Model Context Protocols (MCP), presents a more pragmatic and potentially transformative avenue for blockchain projects seeking to leverage the power of AI. MCPs, by design, facilitate the direct and contextual connection of diverse blockchain-related data sources to LLMs, enabling a new generation of intelligent applications.

MCP Infrastructure Class

This category focuses on building the foundational infrastructure necessary for the effective operation of Model Context Protocols.

• MCP Cloud Service: This direction involves the provision of cloud computing resources specifically optimized for the computational demands of MCP-driven applications. While the underlying technology of cloud services is well-established, projects in this area often integrate specialized features to enhance their narrative and utility within the blockchain context. For instance, DARK's integration of Trusted Execution Environments (TEEs) offers a compelling security enhancement for sensitive blockchain operations within the cloud environment. However, the long-term differentiation and competitive advantage of such projects may be susceptible to advancements and offerings from larger cloud infrastructure providers.
• MCP Tool Platform: These projects aim to monetize the wealth of blockchain data by developing user-facing tools powered by MCPs. Two primary models can be identified:

• Open Platforms: Mirroring the approach of Dune Analytics, these platforms provide users with comprehensive blockchain data and intuitive visual interfaces, empowering them to develop their own customized MCP-powered analytical tools and dashboards. For open platforms, the breadth, depth, and real-time nature of the data, coupled with the activity and expertise of the user and developer community, are critical factors in establishing a sustainable competitive moat. Dune Analytics' potential expansion into MCP-powered analytics could pose a significant barrier to entry for new competitors due to its established user base and extensive data resources. SkyAI's reportedly developing "Playground" may fall into this category, although specific details remain unconfirmed.
• Closed Platforms: These platforms leverage the proprietary data access and rapid development capabilities of their internal teams to build a curated suite of high-quality MCP-driven tools, as exemplified by Heurist AI's MCP toolset. While this model may not foster the same open ecosystem effect as platforms like Dune, it offers the advantage of guaranteed tool quality and seamless integration with the platform provider's specific data sources.

MCP Ecosystem Spotlight: Heurist Agent Framework and Mesh

The Heurist Agent Framework stands out as a versatile multi-interface AI agent framework, enabling seamless interaction across a diverse range of platforms including Telegram, Discord, Twitter, Farcaster, REST APIs, and, critically, Model Context Protocols (MCP).

A recent significant development is the launch of Heurist Mesh, a decentralized and open network where specialized AI agents are contributed by the community and can be utilized in a modular fashion. Each agent within the Mesh represents a distinct functional unit capable of processing specific types of data, generating customized reports, or executing predefined actions, collectively forming an intelligent swarm capable of tackling complex, multi-faceted tasks. Importantly, all Heurist Mesh agents are designed to be accessible via MCP, allowing users to interact with them through their preferred MCP-compatible client applications, such as Claude Desktop, Cursor, and Windsurf.

The Heurist Mesh currently comprises over 30 specialized agents, including notable examples such as:

• BitquerySolanaTokenInfoAgent - Provides comprehensive analysis of Solana tokens including metrics, holders, trading activity, and trending token discovery
• CoinGeckoTokenInfoAgent - Fetches token information, market data, trending coins, and category data from CoinGecko
• DexScreenerTokenInfoAgent - Fetches real-time DEX trading data and token information across multiple chains
• ElfaTwitterIntelligenceAgent - Analyzes tokens, topics or Twitter accounts using Twitter data, highlighting smart influencers
• It is crucial to recognize that within the domain of MCP infrastructure, the accessibility, quality, and contextual relevance of the underlying data often supersede the novelty of the core MCP technology (such as prompt engineering techniques). This presents a significant opportunity for established blockchain data service providers to evolve their offerings and capitalize on the burgeoning demand for MCP-enabled solutions.

Source: Heurist

Application Class

Current Web3 AI applications often function as intermediary layers or "wrappers" around existing Web2 AI products. Typically, significant speculative interest can emerge around Web3 projects that integrate the technologies of leading Web2 AI companies following major product announcements. The increasing sophistication of AI Agent frameworks, such as OpenAI's Agent SDK (encompassing security protocols, management interfaces, permission controls, and inter-agent communication modules), is facilitating the creation of more complex and autonomous agents leveraging diverse toolsets.

The advancement and widespread adoption of robust Model Context Protocol (MCP) infrastructure are expected to significantly accelerate the development of the AI application layer within the Web3 ecosystem. By providing AI agents with contextual awareness of blockchain data, MCPs will enable a new generation of intelligent applications. Furthermore, application-layer products inherently possess greater ease of dissemination and user adoption, contributing to stronger market narratives. Two application scenarios with high potential include:

• Customized Intelligent Assistants: By integrating fundamental MCP functionalities, users can construct highly personalized intelligent assistants tailored to their specific data needs and preferences within the blockchain domain. Future advancements in Planner Agents could even facilitate collaborative workflows between multiple intelligent assistants (Agent-to-Agent interactions) orchestrated through MCPs.
• Workflow Generators: Users can leverage various MCPs and specialized functional agents to design and automate complex workflows involving blockchain data analysis, trading strategies, and other on-chain activities. Compared to general-purpose intelligent assistants, the invocation of MCPs within workflows typically follows a more defined sequence or parallel structure to achieve specific, predefined objectives and deliver actionable results to the user. This type of application is particularly well-suited for tasks with existing mature processes that can be augmented by AI-driven data analysis.

Conclusion

The convergence of AI Agents and cryptocurrency represents a dynamic and rapidly evolving frontier with the potential to fundamentally reshape the Web3 landscape. While the market may have experienced periods of consolidation, the underlying technological advancements and the innovative approaches of projects like the Virtuals Protocol and the ElizaOS platform underscore the significant potential of this synergy. The Virtuals Protocol's emphasis on establishing a robust Agent Commerce Protocol and fostering a contribution-based token launch mechanism presents a compelling model for cultivating an autonomous agent economy. Conversely, while Eliza (Auto.fun) introduces novel features for agent tokenization on Solana, its current iteration appears to lack significant differentiation from existing launchpad models.

Looking ahead, the development and widespread adoption of robust Model Context Protocol (MCP) infrastructure are poised to be the critical catalyst for unlocking the true transformative potential of AI applications within the Web3 ecosystem. By effectively bridging the gap between the vast and complex world of blockchain data and the analytical and generative capabilities of Large Language Models, MCPs might empower a new generation of intelligent tools and applications, ranging from highly customized intelligent assistants to sophisticated automated workflow generators.