In recent months, more and more people have started talking about: agency economy “Agent economy” in Spanish. The term was popularized by a group of researchers at Microsoft in a book published in May 2025 titled “Agentic Economy.”
Here, the authors have outlined how they expect this new economy, which is already underway, to work.
New economic agents: what is it?
To understand economics, we first need to understand what an AI agent is. In this context, agents are referred to as generative AI-powered tools that can generate actions within a system.
ChatGPT or Gemini is a good language model (LLM, large language model) interact through chat, but they are not agents.
Such an LLM responds to requests (prompt) Text or images from users. That’s why it’s generative. In other words, respond to what is asked of you. When I ask an LLM to search for videos on the Internet with explanations, it will return suggested titles and possibly links. Its capacity is text or graphics.
Although I am an agent, execute. If you ask for a video, Your agent may be able to find a way for you to download it.
In this sense, AI agents do not have human-like agency (they have no will or desire of their own), but they do have the autonomy to achieve desired outcomes. Additionally, agents are often specialized in specific types of tasks, rather than being generalists like LLMs.
Agency economies are formed when these agencies are formed. they begin execute a trade economically autonomous and among themselves. For now, agents are focused on the product provider side, such as customer service or sales roles, but this is rapidly changing with the development of personal agents for users.
For example, agents can access email accounts to send personalized emails to a list of users. LLM no longer requires a script to generate or send text. The agent is You can do both or create processes to automate the tasks.
Another example: A person asks an agent to make a hotel reservation, the agent selects a hotel based on the user’s preferences, contacts the hotel agent to make the reservation, and even prepays for the reservation.
In addition to agents provided as a service by major technology companies (such as Copilot, Google Assistant, and Apple Intelligence), open and autonomous tools have also been developed, such as OpenClaw, which allows users to create their own agents using their own hardware.
For such a process to work, in addition to having agents on both sides, they also need a protocol to allow them to communicate (Google’s A2A or Anthropic’s MCP, and Coinbase’s version thereof). Channel so you can make and confirm your payment.
Error 402 and the role of cryptocurrencies in the agent economy
As you will see below, several companies in the crypto space have warned that these are the best option for AI agents. These are open, native internet networks and protocols, making them more accessible to agents. It is different from traditional banking and closed financial systems.
Based on this, these companies started developing standards and protocols to allow AI agents to use their own Bitcoin and cryptocurrency wallets.
The magic number here is not 21. Shino402. A 402 HTTP status indicates “Payment Required.” This is the error message returned in situations where the server denies access to a resource because payment has not been made.
Since there was no way to integrate the payment system directly into the website, that error status remained unavailable, along with more common error statuses such as 403 “Forbidden” and 404 “Not Found.”
In cryptocurrencies, this HTTP response code can be used as a notification of access conditions that require payment on a web page. Currently, there are already several projects trying to build an agent economy within the HTTP protocol.
L402: Lightning Network for AI Agents
Lightning Labs is developing L402, a standard that makes it easy to exchange web services using payments in Lightning. Web services that implement L402 can put up barriers (paywall) Requires customer to pay for Lightning before granting access.
a paywall Traditionally, it requires that the user has an account on the server, that the server is integrated with some kind of payment system, and that external payment validation must be performed through that provider and associated with the requesting user.
Interfaces are usually comfortable for users, but the systems behind them are complex and have many links that can fail. In L402, Lightning payment itself is token Integration is more fluid because client access is restricted.
Of course, it’s much easier for an AI agent to use a Lightning wallet than a credit card in their name.
x402: Agent with USDC
x402 is another similar protocol developed by Coinbase. It is used with USDC on three blockchains: Aptos, Solana, and Stellar.. This process is similar to L402.
- The client makes an HTTP call to the web server (www.miservicio.com/api/resultado).
- Server responds with HTTP 402: Payment required
- The customer pays and calls again. This time I have attached a signed proof of payment showing that I made the payment.
- The server responds with HTTP 200: OK and provides the requested service
x402 also provides a detection layer called Bazaar. Any server offering any payment service via x402 can expose it and it will be visible through Coinbase or PayAI’s API.
Both services can be used by humans, but their documentation is Shows it’s designed with AI agents in mind And they are integrated with communication protocols, mainly MCP.
MPP and OWS: Simplifying the use of AI wallets
Stripe wasn’t far behind and launched the Machine Payments Protocol, which, as the name suggests, allows machines to make payments to each other. Although it is an open protocol, the implementations provided depend on services such as validation and traceability.
Taking this infrastructure to another level, MoonPay Wallet recently launched the Open Wallet Standard (OWS) in collaboration with the entire digital payments industry, from PayPal to TON to the Ethereum Foundation.
OWS is a unique method that allows AI agents to use digital wallets with the same private keys and also allows them to pass control to another agent. The problem this solves is that without it the agent would have to have a specific library for each network it wants to use.: one for Bitcoin, one for Ethereum, one for Solana, etc.
OWS allows agents to create and store private keys and obtain wallets and addresses on any network, allowing them to switch between wallets and addresses depending on the payment required.
Identity and reputation of economic actors
The final pillar needed for this new power relationship is the identity of these agents. They must be able to identify each other and know if they have interacted before. Know if your trading partners are interacting with others and can be trusted; Find out what company or organization you work for.
For all this, several standards and protocols have been developed. For example, Ethereum created and approved ERC-8004 to directly address this issue within the blockchain.
VeryAI, used by Solana, Polychain, and others, uses palm scanning to verify and associate agents with humans. I myself am collaborating on a project called Observer Protocol that addresses this issue.
The project’s founder, Boyd Cohen, also wrote a book titled “Bitcoin Singularity” about the interaction between AI and cryptocurrencies.
The last similar project funded and privately licensed by Ripple was called t54 and was intended to be developed as a service rather than an open standard.
The future of the digital economy
Everything described above gives us an idea of what will happen in the coming months and years: Artificial Intelligence Agents with Autonomybuy and sell services to each other over the Internet.
A personal assistant that you can talk to and ask questions through instant messaging applications buy, sell, reserve, rent. Meanwhile, another agent created by artificial intelligence working for the company assists the assistant and buys, sells, and rents items from the assistant.
They pay each other in cryptocurrencies and provide encrypted receipts indicating that the transaction is complete. They advertise their services through protocols.
Reputation is calculated algorithmically and trust is encoded into interactions and protocols. This is how we describe the “new era” of the digital economy.
(Tag to translate) Cryptocurrency
