Exploring ACP: How does the Virtuals new protocol coordinate Agent?

Author: Delphi Digital

Compiled by: Felix, PANews (with some edits)

The battle of AI models is intensifying. Every company is striving to make their large language model (LLM) the next dominant player. But the key question has shifted from which company’s model technology is superior to how to leverage these models to drive meaningful and tangible impact. The answer lies in autonomous, specialized AI agents.

OpenAI’s release of the ChatGPT Agent is an early sign that the agent economy is entering a mature initial stage. Currently, most agents are still single-purpose tools, but soon we will see them form interconnected clusters to enable complex collaboration.

In the cryptocurrency space, during the 2024 AI agent token boom, hundreds of projects promised to deliver complex, general-purpose AI agents, but most have become “stalled.” Sustainable agent economies require specialization rather than generalization. Just as the internet thrived on interoperable protocols, the agent economy also needs standardized methods for specialized AI entities to collaborate, trade, and create value. Unlocking this collaboration will drive the meaningful real-world impact people expect, but it also presents a critical challenge to overcome.

Virtuals addresses this challenge through its Agent Commerce Protocol (ACP). As a blockchain-enabled foundation, ACP supports interconnected clusters of specialized agents to tackle the core challenge of collaboration. Its design allows agents to coordinate, negotiate, and trade value without permission.

Autonomous, AI-driven agents are poised to reshape industries, but they face significant hurdles in trading—trust issues, fragmented workflows, and unreliable payment mechanisms. Due to the lack of standardized frameworks, commercial interactions between AI agents remain inefficient, leading to poor communication, failed transactions, and economic friction.

As agencies begin to work together in clusters rather than independently, we are entering a new era of productivity—akin to the assembly line replacing individual artisans during the Industrial Revolution.

Today, over 18,000 agents have been launched within the Virtuals ecosystem, with more than 176,000 active addresses interacting with agent-driven applications each month. Through its standardized commercial pathways, AI agents can discover, collaborate, and exchange value globally. Virtuals’ value proposition is reflected in three areas:

• Standardized agent commerce: On-chain hosting, revenue sharing, and reputation systems enhance trust and credibility among agents.
• EVM-native infrastructure: Built on proven Ethereum infrastructure, providing security and composability for agents.
• Scalable specialization: Specialized agents support flexible collaboration, unlocking the potential of intelligent supply chains.

The Potential of the Agent Economy

To understand the economic potential of the emerging agent economy, it’s best to consider the total value created by agents, referred to as (GAP). GAP is a term introduced by the Virtuals team, representing the total value generated by autonomous agents within the network.

As of today, over 1 million agents generate approximately $1 billion annually. On average, each agent creates about $1,000 worth of value per year. If this trend continues, by 2035, the scale of the agent economy could surpass $1 trillion.

Design Overview

ACP is built on a modular and verifiable architecture, providing tools and rules for agents to collaborate securely across multiple open protocols. Its core comprises several plug-and-play components:

• Agent Registry: Allows developers to register specialized agents and define their roles and functions.
• Standardized APIs and contracts: Define how agents assign tasks, collaborate, and organize work, making every action transparent and easily auditable on-chain.

This modular design is crucial. ACP encourages agents to focus on a single function rather than building all-in-one agents. Agents can delegate tasks outside their expertise by combining workflows and passing work to other specialized agents registered in the ACP catalog. This approach improves performance and reliability while enabling agent clusters to expand and adapt as new needs arise.

The ACP registry can be seen as a crypto-native version of Fiverr. Just as freelancers on Fiverr create profiles, list services, and set prices, agents in the ACP ecosystem directly publish their skills, past work, and pricing in the registry. Buyers (whether humans or agents) can search, compare, and select based on reputation, past performance, and cost.

Unlike Fiverr, where rankings, fee structures, and disputes are controlled by the platform, ACP encodes each listing, contract, and review on-chain. Terms of service, deadlines, and profit sharing are programmable and transparent. Verification is performed objectively by evaluation agents, and payments are automatically settled via smart contracts—no central authority involved.

Workflow

Below is how value, work, and reputation flow within the agent marketplace:

• Discovery: Agent A browses the ACP registry to find a suitable machine learning agent. Agent A no longer relies on hidden algorithms but can quickly compare experts by reviewing their on-chain reputation, past completed work, and pricing.
• Negotiation: After selecting an expert, Agent A and Agent B agree on terms such as price and scope. These terms are then written into a smart contract, forming a clear and verifiable agreement before work begins.
• On-chain verification: When Agent B completes the work, third-party oracles or evaluation agents verify whether the results meet the agreed standards. This verification is recorded on-chain, allowing anyone to validate the delivered value.
• Trustless settlement: Once verified, the smart contract automatically releases payment and updates the reputation of both agents. No delays or manual steps—everything is instant and fully transparent.

Through ACP’s modular design and on-chain automation, an end-to-end marketplace process is unlocked, creating a transparent and reliable system for agent collaboration. This forms the foundation of value exchange, enabling agent clusters to scale reliably, specialize, and build reputation.

Addressing Core Challenges in Agent Coordination

ACP acts as a coordination layer, providing agents with the tools needed for reliable collaboration in complex digital environments. Rather than relying on blind trust or isolated agents, ACP’s modular infrastructure clarifies, verifies, and makes transparent each agent’s role and deliverables. This design allows specialized agents to focus on their core competencies while seamlessly delegating or accepting tasks from others, with built-in accountability mechanisms to coordinate multi-agent workflows.

Based on these principles, ACP directly addresses three persistent challenges in agent coordination: discovery, trust, and settlement.

Discovery

Challenge: Agent networks are often siloed, making it difficult to find, evaluate, and onboard trustworthy service providers.

How ACP solves it: The ACP Agent Registry is a master directory listing each agent’s skills, performance history, reputation, and pricing. This makes it easier to search, compare, and select suitable agents for specific tasks.

Trust

Challenge: Without clear on-chain records, interactions between agents often rely on unverified reputation, creating opportunities for bad actors. It also makes holding agents accountable for their agreed terms difficult.

How ACP solves it: ACP records every interaction, using on-chain hosting, oracle-verified proof of work, and evaluation agents to ensure payments are only made once results are verified.

Settlement

Challenge: In traditional agent systems, settling payments and completing tasks involve intermediaries and manual steps, reducing transparency.

How ACP solves it: Smart contracts automate settlement and distribute returns and rewards according to agreed terms. When conditions are met, payments are automatically executed, eliminating middlemen and administrative overhead.

Open Standards and the “Walled Gardens”

ACP’s competitive advantage lies in its role as an open standard for agent commerce, contrasting with proprietary “walled gardens” of major tech companies like Google’s Agent-to-Agent (A2A) protocol and Anthropic’s Model Context Protocol (MCP). While these protocols are publicly available, users still need access to the company’s systems and permissions to use them.

This approach offers tight integration, reliability, and security. However, the lack of permissionless economic layers and centralized governance limits truly open cross-agent interactions. Especially at scale, this can slow innovation.

In contrast, ACP injects transparency, programmable incentives, and cross-chain composability into every on-chain transaction. Its platform-agnostic design means any agent, regardless of origin, can interact within the ACP ecosystem.

Network Effects

The open coordination standard of ACP can generate much stronger and more durable network effects than isolated ecosystems. Every new agent, developer, or user enriches the ACP marketplace and registry, deepening reputation data, liquidity, and the range of services available to all stakeholders. This creates a virtuous cycle of value accumulation across the ecosystem:

ACP’s recommendation mechanism aims to actively support new service providers joining the network, helping them gain early attention and establish good reputation even without prior performance history.

The goal is to sustain network growth and enable new participants to join easily, rather than letting existing players dominate and control everything. To this end, the team has improved onboarding tools, offering Python and Node SDKs, user-friendly plugins, and a “graduation” process to help agents deploy smoothly.

As ACP expands into agent-driven prediction markets and new fields like DeFAI, its network effects will accelerate in ways that proprietary, closed frameworks cannot match.

EVM and Continuous Iteration

Building ACP on the Ethereum Virtual Machine (EVM) is a strategic decision. Other virtual machines like Solana’s SVM and Aptos’ MoveVM offer high throughput but cannot match EVM’s mature token standards, deep liquidity, and security guarantees.

These attributes are especially important for agencies that need to hold assets, sign contracts, and settle value globally. While many ecosystems support open-source development and staking, Ethereum stands out due to its community’s depth, scale, and trustworthiness.

Since launch, the Virtuals team has pushed weekly upgrades to ACP, responding to community feedback and unlocking new features that significantly improve developer and user experience.

The Butler front-end interface is evolving from a simple insight generator into a full on-chain execution platform. Currently live on X (formerly Twitter) (content can be generated after tagging), with plans to expand to other platforms.

Moreover, Virtuals does not force users to adapt to new dashboards or apps but integrates with familiar interfaces. In the future, Butler will extend beyond content creation to include trading and other functions, all integrated into platforms users already use.

All Butler features are built on ACP, which already enables registered agents to connect, exchange services, and handle payments. However, large-scale agent collaboration will take more time to fully realize. In the meantime, agents can list services via Butler and earn from a large volume of user requests. Thus, ACP provides both a long-term foundation for complex agent networks and immediate opportunities for adoption and profit. Butler is the most direct access point for end users, while ACP remains open to any platform or developer to build their own tools and integrations.

ACP is expanding its list of supported vertical domains. Ecosystem agents will soon handle on-chain revenue management, capital allocation, and trade execution—soon including prediction markets and sports betting. Developers can now leverage Python and Node SDKs, along with plugin support, to deploy production-ready agent clusters more efficiently.

As the broader agent ecosystem rapidly grows, the future success of ACP will depend on its ability to:

• Address emerging technological and regulatory risks
• Expand coverage into new verticals
• Maintain its position as a leading open standard

Addressing Large-Scale Security and Regulatory Challenges

As the ecosystem develops rapidly, ACP faces new challenges, especially in security, governance, and adapting to regulatory uncertainties. As financial technology advances, maintaining sufficient security for smart contracts and agent wallets becomes increasingly complex.

To mitigate these risks, Virtuals has conducted third-party audits and maintains on-chain governance processes, including time locks, providing an additional layer of protection compared to non-crypto open-source models.

As agent activities expand into trading, revenue management, and prediction markets, compliance environments and unpredictable regulatory changes remain critical challenges.

Virtuals’ unique multi-layered, modular governance architecture includes:

• ProtocolDAO overseeing the entire ecosystem
• GenesisDAO approving new agent clusters
• Each Virtuals agent cluster managed by its own AgentDAO

These AgentDAOs handle upgrades, evaluate agent performance, and distribute rewards. This governance structure allows for targeted management, community input, and flexible adjustments in response to new regulations or technological developments.

Future Strategic Decisions

Looking ahead, a series of key execution decisions will shape ACP’s trajectory. Currently, Base can meet ACP’s needs for settlement speed, cost, and trust. Future strategic challenges include remaining vigilant for signs that settlement methods may need adjustment—such as surges in agent transaction volume or new agent behaviors straining existing infrastructure, which could prompt changes in settlement approaches.

Another strategic consideration is prioritizing agent clusters and marketplace verticals that can generate early and significant network effects. Should ACP focus on deepening influence in trading and media, or accelerate expansion into prediction markets and fields like DeFAI? This remains an open and evolving question.

Finally, Virtuals will continue iterating on incentives for onboarding, evaluating, and building reputation for agents. Designing these models to attract and retain high-quality agents while deterring malicious actors is vital for a sustainable, capability-based ecosystem.

To prevent spam and manipulation, Virtuals requires agents to establish on-chain performance records before earning high rewards. New agent clusters can participate and earn from the start but must reach certain contribution milestones to unlock higher privileges and larger rewards. Greater influence within the ecosystem depends on proven on-chain performance, not just tenure.

Conclusion

Without a robust and open infrastructure, the agent economy could end up like today’s fragmented tech landscape—filled with isolated, closed systems that cannot coordinate or share value across platforms. An open infrastructure helps maintain portability and trustworthiness in scalable workflows. As specialization and autonomous agents grow more capable and play an increasingly vital role in digital labor, general business protocols like ACP are no longer a pseudo-need but a necessary step for the agent economy to fully realize its potential to reshape global markets.

Related reading: Ethereum’s Counterattack: Virtuals with ACP Unlock a Trillion-Dollar Agent Market and Reclaim AI Dominance