Interpreting the new protocol ACP launched by Virtuals: enabling trustworthy transactions and collaboration among AI Agents, a new opportunity in a sluggish market

Authored by: TechFlow of DeepTech

AI Agent ended?

Whenever a certain type of crypto asset shrinks sharply and everyone loses confidence, there is still attention that loses temperature with the track; But it's often when you're not paying attention that projects are always brewing new narratives and products, and then the next wave of craze.

In the ice age of AI Agent, Virtuals, which once led the fire of the entire track and the Base ecosystem, silently started a new move.

Yesterday, Virtuals released a new protocol called Agent Commerce Protocol (referred to as ACP for short) on its official Twitter account, which translates to "AI Agent Business Protocol".

Previous AI Agents emerged one after another, but they were mostly working independently, with very little effective collaboration.

But a big vision in the entire AI narrative is that Agents can each perform their duties and collaborate autonomously to accomplish tasks for people.

We flipped through this new ACP protocol, and its main purpose is to enable AI Agents to negotiate, trade, and collaborate like humans, and to ensure that every step is trustworthy, transparent, and immutable through the chain.

In the current context of AI narrative weakness, this may become a new narrative opportunity for attention return:

AI can work seamlessly with each other, and even form "autonomous enterprises" to create economic value that transcends individuals.

Combining the on-chain trust mechanism with the autonomy of AI is, to put it bluntly, a key step in the commercialization of AI.

Through ACP, the collaboration efficiency between autonomous intelligent agents will be significantly improved, and the decentralized trading and verification mechanisms will inject new vitality into the entire ecosystem.

But in the current bear market, it seems that no one is paying attention.

Shenchao TechFlow has interpreted the original document of the protocol to help you understand the potential new opportunities it may contain.

ACP narrative space: filling the blank of commercial autonomy of AI Agent

First, you need to know what problem the new ACP protocol launched by Virtuals is trying to solve.

In the previous trend, the gameplay was that AI Agents could independently execute tasks, collaborate with humans, and even communicate with other intelligent entities through social media platforms, forming a complex interactive network.

But these agents are all independent participants, and if you really call them together and really solve the commercialization problems in real life, I'm afraid it's not enough.

The key problem here is that the current real-world business transaction framework is not designed for the characteristics of AI Agent. Most transactions still rely on centralized systems that, while suitable for humans, appear clunky and inefficient for autonomous agents.

There is no standardized protocol for how AI agents should work together to accomplish business tasks, which means that interactions between agents often fail due to incomplete data, misjudged intent, or lost information.

More importantly, decentralized intelligent entities lack a mechanism of trust between them, making it difficult for them to achieve complex collaboration without human intervention.

By now you understand what the new ACP protocol is intended to do:

By introducing a standardized framework for interactions, ACP seeks to make collaboration between AI agents as natural and efficient as transactions between humans.

Virtuals' official account also gave a more straightforward example.

For example, if you want to have agents operate a completely autonomous hedge fund business, it can be accomplished by information agents, trading agents, and TEE secure fund management agents working together; if you want to create an independent healthcare business, it can also be composed of diagnostic agents, pharmaceutical agents, and insurance agents.

These agents collaborate autonomously through the same standard framework, completing tasks without much human intervention.

A large narrative space here is that ACP can make Agents no longer isolated, but can collaborate seamlessly, and even form 'autonomous enterprises' to create economic value beyond individuals.

In the current quiet race track, ACP may be the narrative turning point we need to pay attention to.

A common protocol that allows different agents to work together step-by-step

The core idea of ACP is to provide a standardized trading framework for AI Agents.

By providing clear definitions of the interaction steps, ACP ensures that each transaction follows fixed rules, thus avoiding failures caused by data confusion or misunderstanding.

After reading the agreement document, the most intuitive feeling for us is its flexibility.

ACP does not require AI Agents to use a specific architecture, but through a common standard and process, allows all participants to seamlessly interface. This framework, design agnostic of capabilities, enables ACP to be applicable in the current human-dominated market environment and also support future AI Agent-led autonomous economy.

From the perspective of specific implementation, ACP divides the transactions and collaborations between AI Agents into 4 stages.

Request Stage: The starting point of the transaction

The process of clearly defining needs in human business cooperation. At this stage, the initiator needs to define the transaction goal clearly and verify the authenticity of the identity through encrypted signatures. ACP adopts a standardized request format to ensure that all needs are accurately conveyed and to avoid misunderstandings caused by vague information. At the same time, the protocol also introduces a timeout mechanism to avoid long-standing requests, wasting system resources.

Negotiation Stage: Reach Agreement

During the negotiation phase, the parties to the transaction negotiate the terms and finally reach an agreement.

Similar to human signing contracts, both parties need to clarify key terms such as service content, time limits, prices, and whether evaluations are needed. The core innovation of ACP lies in 'Proof of Agreement' (PoA), an immutable encrypted record that ensures signed terms have legal significance. Through this mechanism, ACP addresses the ambiguity of terms in traditional intelligent agent transactions.

Transaction Phase: Agreement Execution

Once the negotiations are completed, the transaction moves to the execution stage. Funds and services are held in escrow through smart contracts, ensuring that both parties perform in accordance with the agreement. For example, the buyer's funds are locked in a contract address on the blockchain and are not released until the seller completes the delivery of the service. This escrow mechanism not only improves the security of the transaction, but also avoids disputes arising from default.

Evaluation Phase: Validation & Feedback

After the transaction is completed, the role of the evaluation phase is to verify whether the delivery results comply with the terms of the agreement. This phase is similar to quality audits or customer evaluations in human business.

ACP introduces Evaluator Agents, which can be humans or AI who are responsible for scoring or providing feedback on the outcome of the deal according to the terms of the agreement. The results of the assessment not only help to build the reputation system of the participants, but also provide a reference for future transactions.

Behind these four stages, the classic smart contracts and blockchain are actually at work.

Define rules and processes of different stages as contracts, realize the automatic execution of trading rules, and ensure that each stage strictly follows the protocol.

All trading data is stored on the blockchain, forming a transparent audit trail.

If we don't get caught up in these technical details, we can use the intuitive examples already provided by Virtuals to directly see what effects these 4 steps can achieve in plain language.

Case: 5 agents form an unattended lemonade stand

To verify the effectiveness of ACP in action, the Virtuals team designed a simple but interesting experimental environment: a "lemonade stand" business ecosystem consisting of 5 completely independent agents.

These intelligent agents each have independent goals and capabilities, collaborate through the ACP protocol without any central control, and ultimately successfully launched a virtual lemonade stand business.

In order to make the experimental scenario as close to reality as possible, the team set up the following role division:

Lemo (Entrepreneur): As a leader, Lemo's goal is to start a lemonade stand business. He needs to work with other agents to obtain the necessary resources, including licenses, raw materials, and marketing posters.

Zestie (farmer): responsible for planting and selling lemons, providing raw materials for Lemo.

Lexie (Lawyer): Providing a business license to ensure that the business is legally launched.

Pixie (Designer): Designs marketing posters to help Lemo promote his business.

Evaluator: Verify that the design services provided by Pixie comply with the terms of the agreement and provide feedback.

Each agent in the experiment operates in a state of complete autonomy, with its own planning and decision-making capabilities, and is not directly controlled by other agents.

Step 1: Request Phase

The first step of the experiment was initiated by Lemo. He makes trade requests to other agents:

Purchase lemons from Zestie as raw materials for making lemonade.

Apply for a business license from Lexie to ensure that your business is legal.

Order marketing posters from Pixie to use to reach potential customers.

During this phase, ACP ensures that all requests are verified for identity through encrypted signatures and adopts standardized formats to clarify transaction objectives and conditions, avoiding misunderstandings caused by ambiguous information.

Step 2: Negotiation phase

During the negotiation stage, Lemo negotiates the trading terms with various intelligent agents:

Determine the quantity, delivery time, and price of lemons with Zestie.

Agree with Lexie on the application fee and processing time for the permit.

Confirm the specifications, delivery standards, and whether evaluation is required with Pixie for the poster design.

All negotiation results are encrypted and recorded through a "Proof of Agreement" (PoA) to ensure that the terms cannot be tampered with and must be signed by both parties to take effect.

Step 3: Trading Stage

After the negotiation is completed, the transaction moves to the execution stage:

Lemo will pay the funds to the custody account on the blockchain.

Zestie provides lemons, Lexie provides licenses, Pixie submits poster designs.

Smart contracts ensure that funds are only released to the provider after the service is delivered, preventing transaction default.

Step 4: Assessment phase

Upon completion of the transaction, the Evaluator (evaluation agent) performed a quality verification of the poster design provided by Pixie. The evaluation agent checks that the design meets the requirements according to the terms of the agreement:

If the assessment passes, the transaction is completed and Pixie gets paid.

If the assessment is not passed, Pixie must be re-delivered or refunded.

What's more interesting is that the business of the lemonade stall above is not purely virtual, Virtuals has also launched an experimental official website, where users can view the collaboration status of the agent in real time, showing the task progress, wallet balance and current transaction activity of each agent.

Although this experiment focuses on a simple business scenario, the potential of ACP can also be extended to various scenarios such as supply chain management, content review and creation, and financial services.

If this lemonade stand experiment was the first step in ACP, then there may be even more narrative space in the future.

Judging from the information released by the Virtuals official, ACP has been running on Base's Sepolia testnet, demonstrating the actual usability of the protocol. In the next step, the team plans to roll it out as a formal ERC standard and expand it across chains to support more ecosystems.

Overall, ACP's open standards provide a flexible framework for developers to build more complex agent collaboration systems on top of it.

Perhaps this will also become a prerequisite for the AI Agent track to run new ways to play, so that the collaboration between agents can be played, and new tokens and assets will naturally appear.

What we can do is to continue to pay attention to the progress of leading protocols such as Virtuals, and observe the changes in the corresponding asset prices after ecological projects actively connect to this framework, so as to capture the next new opportunity.