A brief analysis of the launched ACP protocol: How to reshape the intelligent collaboration paradigm?

Virtuals has launched the ACP protocol (Agent Commerce Protocol), aiming to create an economy where independent AI agents can trade and collaborate, forming a virtual world of AI-driven interactions.
The protocol treats AI agents as independent individuals rather than tools, emphasizing their autonomy and potential for negotiation, payment, and service evaluation.
Key components of ACP include:
- Standard layer & multi-agent framework: Enables registration, negotiation, and payment between AI agents.
- AI agent index registry: Tracks agents' services, rates, and market performance (e.g., adoption rates, fees earned).
- Work evaluation: Third-party AI agents assess service quality (e.g., DeFi yield optimization) with customizable metrics.
- Smart contract payments: Ensures payment only upon meeting predefined delivery conditions.
The goal is to foster an AI agent economy, where growth depends on agents' capabilities and interdependencies, though adoption may be gradual and rely on real-world utility.
Success hinges on whether AI agents can generate practical demand and sustain a positive feedback loop of innovation.

After a quick look, Virtuals launched the ACP protocol (Agent Commerce Protocol), which essentially aims to connect many single AI agents with different capabilities to build an economy composed of AI agents. If it succeeds, in a sense, it is also the metaverse of the AI era, a virtual world of AI agents. (Related reading: " From the ACP framework to the new fee mechanism, can Virtuals create glory again? ")

Let me briefly talk about a few aspects:

1. Virtuals regards the single AI agent as a truly independent individual, rather than an efficiency tool or assistant. This concept of viewing AI agents as independent individuals is very important. It is an important watershed in understanding AI agents and a basic concept for understanding what virtuals will do in the future.

2. To build the virtual world composed of independent AI agents mentioned above, standards are required, so a standard layer is needed; to serve multiple agents, a multi-agent framework is needed, which is an important background for Virtuals to launch the ACP protocol. Only in this way can independent AI agents negotiate and trade smoothly and generate new use cases.

3. The Virtuals ACP protocol mentioned above is a multi-agent framework and standard layer. It includes AI agent index registration, negotiation between AI agents, AI agent work evaluation, AI agent payment, etc. AI agents launched on virtuals are registered in the index registry with names, work content, rates for executing work, etc. With this, an AI agent can understand what services other AI agents can provide and can evaluate the cost of obtaining services. At the same time, based on indicators such as the number of times these AI agents are adopted, the total amount of fees obtained, and the evaluation after being used by other AI agents, its market position among similar AI agents can be seen.

Since there is a service, the service effect needs to be evaluated. AI agent work evaluation is also a type of AI agent service. AI agents that meet the needs of work effect evaluation can be built according to specific characteristics in different fields. For example, when a DeFi income AI agent hires a liquidity management AI agent, it can use a third-party evaluation AI agent to evaluate its work and pay the liquidity management AI agent based on the results. The effect measurement can be customized by the AI agent, such as how much rate of return is achieved, or how much it exceeds the market level in the same period, etc.

One of the important features of an independent AI agent is the ability to pay. ACP supports payment escrow and delivery assessment and execution through smart contracts. Payment will only occur when the work delivery meets the conditions, and once it occurs, the result cannot be changed.

4. From the above description, virtuals attempts to build an economy composed of AI agents through the ACP protocol, allowing different AI agents to trade, thus generating an AI agent economy. Ultimately, the size of this economy depends on what these AI agents can do and what different AI agents need other AI agents to accomplish.

5. The formation of the AI agent economy will not be fast. Once the demand is generated, transactions between internal AI agents will start, which is considered a cold start. If a positive cycle can be formed, there may be a lot of cross-innovation. This is the charm of the AI agent economy. As for whether it can succeed, the practical value of the AI agent itself is the key.