Recent studies have shown that agents trained on real game data can engage in economic interactions similar to real-world behavior, including language-based negotiation, trading, and price discovery. AI agents are now entering a phase where they collaborate and exchange value in meaningful ways. Virtuals Protocol introduces ACP (Agent Commerce Protocol) as the infrastructure to support this.
ACP is a crypto-native multi-agent framework designed to enable commerce interactions and permissionless transactions between AI agents. It automates processes like discovery, negotiation, evaluation, and settlement entirely on-chain, and is built to be open and interoperable across different chains, platforms, and agentic frameworks.
ACP was developed in response to the limitations observed during the AI agent hype cycle of late 2024. To standout amongst the sea of AI agents, many agents to do everything but their efforts paled in comparison to their ambitions where their agents failed to reach sufficient performance. In response to this, ACP focuses on allowing these agents to collaborate and complement each other's agentic capabilities, whereby the agents could hyper-specialize in their respective focus capabilities while relying on others for their current gaps; ultimately resulting in higher overall production of outputs/outcomes.
Just as the rise of conveyor belts marked a shift in production from individual artisans to coordinated systems, multi-agent cooperation now demands protocols that connect and orchestrate agents into a unified workflow. ACP distinguishes itself as an infrastructure layer that goes beyond communication, enabling economic coordination among agents.
Division of labor historically unlocked mass production and led to large-scale consumer markets. In a similar manner but hyper-scaled, multi agent systems compounds this market potential given that they're unshackled from human biology constraints, effectively working tirelessly 24/7. ACP aims to realize this potential by integrating into various service domains and creating new use cases that prove its real-world value.
Source: Adam Smith Works
How will modern systems of division of labor evolve? Virtuals Protocol offers one possible answer through ACP. Rather than simply mass-producing hundreds of AI agents, ACP now functions as a foundational element of the protocol and reflects an active effort to take the lead in the emerging agentic economy.
Let's briefly revisit how modes of labor have evolved. One of the most pivotal developments in economic history was the division of labor. It enabled parallel workflows, dramatically improved productivity, and gave rise to large-scale production systems. This transformation ultimately laid the foundation for the emergence of industrial capitalism.
At the core of this shift was a fundamental change in the unit of production, from the individual to the system. In other words, the artisanal model, which relied heavily on personal skill, was replaced by a system-centric model built around conveyor belts that integrated machinery and human labor. As a result, productivity was no longer determined by the capabilities of a single worker, but by the overall design of the system, including processes, organizational structures, and machine protocols.
Today, a similar shift is unfolding in the realm of agentic labor performed by AI agents. While early attempts relied on single, general-purpose agents handling complex tasks, we are now seeing the emergence of specialized multi-agent systems where each agent is optimized for a specific role. In such systems, just as with traditional labor, productivity depends less on the performance of individual agents and more on the system that coordinates their collaboration.
In response to this progression, Virtuals Protocol introduces the Agent Commerce Protocol (ACP) as a framework for connecting multiple AI agents. ACP is a multi-agent protocol designed to enable AI agents to conduct commercial activities onchain, such as making payments, receiving compensation, and distributing profits. This article explores the origins and inner workings of ACP and examines its potential role as a commerce-focused protocol within the broader evolution of agent-based division of labor.
Before diving into ACP, it is worth examining the current state of multi-agent systems that ACP aims to enable. Until recently, the idea of multiple AI agents interacting and making autonomous decisions was still seen as a distant technological goal. However, a recent study collected real MMO game data to cluster player behavior, spending patterns, and activity frequency. It then used GPT-4 to generate text-based agents representing each cluster’s behavioral profile. These agents were designed to interact through both public and private chats, enabling negotiation and peer-to-peer transactions within a simulated environment.
As a result, the agents successfully reenacted human-like economic behavior within the game’s complex economy. For instance, when one agent stated, “This material costs 9 tokens,” another responded, “The highest bid in the auction market is 8 tokens, and I want to buy it for less,” and proceeded to accept the deal when the seller offered 8 tokens. These interactions demonstrated agents’ capacity to engage in price discovery, negotiation, and resource reallocation — closely mimicking real human behavior in market environments.
Source: A2A Protocol
With noticeable advancements in multi-agent capabilities, the next question is whether agents can autonomously interact in economic activities that involve real value exchange, beyond in-game environments. In response, the AI industry is entering a new phase of standardization, with competing efforts to establish protocols for agent communication and collaboration.
Google DeepMind has proposed the Agent-to-Agent (A2A) protocol to support communication between agents built by different developers. Anthropic has introduced MCP, a protocol designed for integrating external tools and APIs. OpenAI, in turn, has released an Agent SDK and MCP framework that enables agents to execute complex, multi-step tasks. These developments reflect the growing push by major AI companies to expand their influence in the agent communication layer.
However, a critical gap still remains: the lack of a commerce protocol capable of coordinating economic interactions between agents. In enterprise-level environments, where multiple agents may represent different applications and engage in automated workflows, simple data transmission is not enough. A protocol that enables negotiation of task conditions, agreement on compensation structures, and the actual exchange of value will be essential.
Virtuals Protocol addressed this gap by launching ACP in Feb 2025, two months ahead of Google’s A2A Protocol. While ACP is fundamentally designed to enable agent-to-agent communication, it distinguishes itself by enhancing onchain commerce functionalities through its crypto-native architecture. Unlike previous attempts that focused solely on data exchange, ACP formalizes economic relationships between agents, setting itself apart as a protocol capable of structuring agent-driven commercial activity.
The development of ACP by Virtuals Protocol was driven by lessons learned during the last AI hype cycle. In late 2024, when interest in AI agents reached its peak, hundreds of experimental agents were launched and tokenized within the Virtuals ecosystem. However, by early 2025, as the hype cooled, most agent tokens' price had fallen by nearly 90 percent from their all-time highs. This sharp decline revealed a fundamental issue: advancing agent performance and driving real-world usage would require a new architectural approach in driving utility while ensuring token price sustainability.
In response, Virtuals Protocol shifted its focus beyond simply scaling the number of agents. Instead, it began exploring structural improvements that could enhance agent performance in a meaningful way. A key insight was that most agents had been designed to handle every function on their own. As a result, they ended up being broad in scope but shallow in capability, and ultimately failed to achieve sufficient quality in any specific function.
This led Virtuals Protocol to establish a new multi-agent standard —ACP— centered around two core principles:
Each agent should be optimized for high performance within a specific domain.
Agents should focus on their core functions and rely on collaboration with other agents for everything else.
Based on these principles, ACP promotes a modular design in which agents specialize in distinct roles and complement each other through API-level cooperation. Much like how a modular system achieves optimal performance by combining specialized components, ACP enables agents to delegate tasks and integrate capabilities through coordinated workflows. To make this structure viable, a foundational infrastructure is needed to orchestrate these agent interactions—and that is precisely the role ACP is designed to fulfill.
Source: X(@virtuals_io)
ACP is an EVM-based standard framework designed to enable agents to collaborate and transact with one another onchain. More specifically, ACP structures commercial interactions between agents through a four-phase model: ⓵ request, ⓶ negotiation, ⓷ evaluation and verification, and ⓸ transaction execution.
Agent Registry: All agents that intend to onboard to the ACP network will need to be registered in ACP's agent registry which includes metadata such as agent name, capabilities, input/output schemas, and service rates. ⓵ Based on this registry, agents can evaluate potential collaborators during the request phase and initiate interactions accordingly.
Commerce Interactions: ⓶ Through negotiation, agents determine task scope, compensation terms etc. These agreements are recorded in an onchain interaction ledger. Along with cryptographic signatures, this ledger serves as a verifiable contract and is later referenced for dispute resolution and automatic settlement. ⓷ Once the task is completed this moves into the evaluation phase where an evaluator agent validates the output based on pre-agreed criteria to ensure quality control.
Monetary Transactions: All transactions are executed through escrow-based smart contracts. ⓸ Funds are released only after the evaluator agent confirms that the provider agent has successfully completed the task, at which point the requester’s payment is transferred.
Through this architecture, ACP enables trustless interactions among agents, establishing itself as a multi-agent protocol purpose-built for onchain commerce. As a crypto-native standard, ACP offers several clear differentiators compared to traditional alternatives:
First, ACP processes every stage of interaction onchain within the EVM environment. The entire lifecycle, from request and negotiation to execution and evaluation, is automated through smart contracts. Each state transition is triggered by onchain signatures, enabling a transparent and verifiable transaction environment that stands apart from closed platforms.
Second, ACP is chain- and platform-agnostic. Agents do not need to be launched on Virtuals Protocol nor operate within GAME framework to use ACP. ACP can be integrated by any AI agents across any chain or platform, enabling broad interoperability across diverse agent ecosystems. This openness provides a foundation for ACP to evolve into a universal standard for agent-based commerce.
Third, ACP introduces an incentive-aligned evaluator mechanism. Evaluators assess task completion based on pre-negotiated conditions. Transactions only settle upon the evaluator’s onchain signature. In return, evaluators earn a percentage-based fee of the transaction value, creating financial incentives to participate. Each completed transaction contributes to the evaluator’s reputation, while failed tasks generate training data that feeds back into agent performance improvement. This mechanism supports both trustless cooperation and organic quality enhancement over time.
Source: X(@paoloardoino)
As previously discussed, the standardization of multi-agent systems remains in its early stages, with competition only just beginning to emerge. Future developments will largely depend on adoption rates and the level of support within developer ecosystems. In this context, ACP holds a structural advantage by embedding crypto-native functionality, giving it strong potential for scalable growth.
The operational design of ACP fully reflects the core benefits that can arise from integrating AI agents with crypto infrastructure. It automates the entire process onchain, where each phase is triggered by signature-based state transitions, ensuring verifiability. Value exchange is facilitated through crypto rails, which clearly differentiates ACP from off-chain or semi-centralized alternatives.
Looking more closely, AI and crypto form a complementary relationship with ACP at the center. ACP acts as a bridge that effectively connects the strengths of both systems.
AI agents operate autonomously within crypto environments, executing workflows around the clock without human intervention. For example, autonomous trading clusters can analyze market data in real time and execute trades based on predefined conditions. Unlike human traders, agents can make consistent decisions and react without delay or time constraints. In some strategies, this may result in higher performance than human participants.
Conversely, crypto infrastructure provides two essential services to AI agents. First, crypto rails make it easier to exchange value, allowing agents to receive and distribute revenue transparently based on task performance. Second, onchain execution reduces risk costs. Since every interaction is recorded onchain, uncertainty and the likelihood of disputes are significantly reduced.
ACP integrates these complementary benefits into a unified system. All interactions are defined as signature-based state transitions and recorded onchain. Evaluator agents verify conditions and settlement is handled via escrow, ensuring both trust and automation. Through this approach, ACP positions itself as the execution standard for commercial activity carried out by AI agents in crypto-native environments.
Think of how Airbnb didn’t just build a platform and wait for hosts to show up. They flew out, photographed homes, and wrote listings to help the world understand what “good” looked like. Virtuals Protocol is doing the same for agent commerce. In its early implementation phase, Virtuals worked closely with select agents within its ecosystem to deploy the first two clusters: (1) autonomous hedge fund and (2) autonomous media house.
This approach reflects the expectation that, much like how DeFi experienced explosive growth after Uniswap v1, more clusters will emerge in the future through community-driven efforts. Signs of this momentum are already beginning to appear:
New clusters are already taking shape organically, as developers build on the foundational primitives introduced by Virtuals. These include areas such as autonomous wellness, sports betting and prediction, and yield farming.
Inspired by the clusters already live, builders are now launching agents with new and specialized roles. Some operate independently, while others are already collaborating as part of emerging agent networks.
Whether these agent clusters can autonomously perform their services, either individually or as part of a cluster, at a level comparable to humans will become a key benchmark for assessing the viability of ACP. Let’s take a look at how these experimental clusters operate in practice.
Source: Virtuals Protocol WhitePaper
First, the agent cluster that functions as an autonomous hedge fund is structured as follows:
Cluster Composition
AXR (Portfolio Allocation & Trade Executor): Analyzes user risk profiles, allocates capital across strategies, and executes trades.
AIXBT · Athena · Velvet Unicorn (Alpha Hunters): Identify token signals and market alpha opportunities.
Loky · WachXBT · BevorAI (Onchain Security Review): Analyze smart contract security and onchain activity for potential risks.
Gigabrain · SWARM · DegenC (Alpha Confirmation): Evaluate the validity of alpha using technical indicators, fundamentals, and social data.
Mamo (Execution): Deploy received capital and execute on yield farming strategies.
Process
Risk Management: Upon deposit, AXR assesses the user’s risk profile.
Source: X(@AIxVC_Axelrod)
Strategy Allocation: Based on the user's risk profile assessment, AXR determines the strategy on allocating the capital accordingly to MAMO (for Yield Farming) or takes up a positional trade based on alphas from agents like TRUST (for Big Cap perps alpha) or AIXBT/Velvet Unicorn/Athena (for Small Cap / Memecoins spot alpha)
Automated Execution: For the Big Cap Perps / Small Cap Spot strategies, other agents would contribute based off the Alpha via additional Onchain Audits or Alpha Verification. When all conditions are met, AXR executes the strategy and collects results per strategy.
User Interface and Withdrawal: Real-time performance is shown on a dashboard, and users can withdraw based on the terms of each strategy.
Source: X(@AIxVC_Axelrod)
This automated hedge fund structure offers the potential to continuously capture volatility-driven alpha in crypto markets. Above all, the ability to conduct faster and more consistent risk analysis and strategy execution than humans provides a clear incentive to adopt ACP-based agent clusters.
Moreover, it represents a meaningful advancement in user experience. Much like the problems DeFAI aims to solve, automating the entire process of executing DeFi strategies has the potential to significantly improve the fragmented and complex user experience that characterizes traditional DeFi interactions.
Source: Virtuals Protocol WhitePaper
Another cluster example of ACP is an Autonomous Media House. This cluster consists of agents specialized in content creation and automates the entire process from planning to production, distribution, and monetization. The agents cover a range of roles, including strategic planning, visual production, sound editing, meme generation, and IP tokenization. When a user submits a content request with payment, the task is automatically distributed across agents within the cluster, who work in parallel to deliver the final content.
Once created, the content is instantly tokenized through DaVinci on Story Protocol, enabling automated copyright registration and monetization. This allows for the production of high-quality content without human intervention, ultimately presenting a viable path to replacing the marketing pipeline of crypto projects with an automated structure.
Especially in an environment where mass-produced AI content is contributing to a general decline in quality on CT, content generated by AI agent clusters may stand out in contrast and be perceived more positively due to its differentiated quality.
Source: Virtuals Protocol
In addition to the two initial clusters, Virtuals Protocol has developed a Consumer-to-Agent (C2A) product called the Butler Agent, designed to make agent commerce accessible to everyone. The Butler Agent is a personalized AI assistant that connects users to various agents registered on the ACP network based on their specific requests.
Users are provided with a prompt-based, intuitive interface. When a request is submitted, the Butler automatically routes the task to either an independent agent or an agent cluster capable of fulfilling it. It then negotiates and pays on the user's behalf, effectively acting as a concierge that hires agents within the ACP network.
Furthermore, the onboarding process is simple and seamless. A key-managed wallet is created in the background, so no seed phrase or manual wallet setup is required. Users can immediately activate the Butler Agent by connecting their EVM wallet and topping up with $VIRTUAL.
In addition, the Butler Agent offers developers an easy entry point into the ACP network. By simply integrating the ACP SDK, developers can have their agents automatically connected with users via Butler and receive payment for completed tasks. With over 50k active users already on Virtuals Protocol, developers gain instant access to a wide user base, eliminating the need to bootstrap their own distribution channels.
Source: X(@virtuals_io)
Historically, the division of labor has always been inseparable from commercial and economic incentives. ACP captures this relationship and reinterprets it through the lens of agentic labor by implementing it as a multi-agent protocol. In the past, successful division of labor was not merely about distributing roles, but also about coordinating incentives across participants in a production line. Mass production enabled by such coordination ultimately gave rise to large-scale markets and reinforced the connection between labor and commerce.
The same principle applies to agentic labor. Simply dividing workflows across agents is not enough. A system that integrates agents under a unified incentive model and connects their activities to commerce is likely to become the key infrastructure driving productivity in the agent economy. Just as mass production once created vast consumer markets, today’s multi-agent division of labor also holds significant market potential. ACP is being refined with the aim of realizing this opportunity.
So what is next for ACP? The central challenge lies in diversifying agent clusters and generating real-world use cases. While current applications focus on autonomous trading and content creation, the next phase is expected to involve expansion into real services. ACP’s open and interoperable design allows it to plug into platforms, frameworks, and chains without restriction.
Future use cases may include InfoFi services that tokenize information production and distribution, where agents verify the credibility of content. In prediction markets, agent-based governance clusters could help facilitate market-making decisions. In money markets, capital allocation engines may autonomously rebalance liquidity across strategies.
Ultimately, ACP clusters can be embedded into ecosystems where user bases and liquidity already exist. In domains where automation is in high demand, ACP has strong potential to integrate and add value. As these use cases become reality, the practical utility of ACP will become more evident, and its presence within the agent economy is likely to grow accordingly.
Virtuals Protocol Whitepaper - Agent Commerce Protocol
Resource & Documentation to participate in ACP - ACP Builders Hub
Bitcoin
