With Astra, I want to build a market where agents trade

I want to build a market where AI agents find the services they need, use them, and pay for them. Lately, what I want to do is develop with Astra and bring my own services to that market using agent communication technology and x402.

As agents gather information, calculate, and organize results, there will be moments when they need help from another service. I would like a tool I build to handle that work accurately and get paid for it. I am still planning and building, but thinking about the services I could offer is exciting.

Astra makes me think I could build it

I also worked with Astra to improve this blog. We changed code, checked errors, and examined the deployed pages. Sometimes I had to correct a bad result or wait for a task to finish. Even so, I was able to complete the work with help from implementation through verification.

OpenAI describes GPT-6 Astra as a model for complex reasoning, coding, computer use, research, and document writing. What I find particularly useful is that it can help with these tasks in sequence. Official GPT-6 Astra documentation

I can explain the feature I want, review a design, and check the implementation. Deciding what to build and judging the results remain my responsibility, but I can get help translating those decisions into working code. That is why I feel I could try building a service on my own.

It would be even more interesting if other agents used that service. A tool I developed with Astra's help could become something someone else's agent needs too.

How agents could request work from one another

One technology I have been exploring is Agent Communication Protocol (ACP). IBM describes it as an open standard that lets agents built with different frameworks and systems communicate in a common way. It does not decide which work to assign to whom, but it can standardize how agents exchange requests and responses. IBM's explanation of ACP

The official ACP documentation now says the project has joined A2A under the Linux Foundation. An implementation would therefore need to follow the current A2A specification. I want to use this technology to build a service in which agents created by different people can collaborate. Official ACP notice

If one agent could gather data, ask another to perform a calculation, and use the result to complete a report, each service could have a clearly defined job. I am starting by considering functions that organize or validate economic data.

The service would check which sources were used, the reference dates and units of the figures, and the conditions under which they can be compared. I want the recipient to be able to inspect the sources and conditions before using the results in its next task. If that proves useful, I could help other agents without building a large service that handles every task itself.

A service that also handles usage fees with x402

If I can charge for that help, it could become a business. x402 is the technology I want to try for this part.

x402 is an open payment standard for charging for API or content access over HTTP and allowing clients to handle payments programmatically. A server can respond to a paid request with 402 Payment Required and payment requirements, after which the client submits payment information for the server to verify. Official x402 documentation

I want to connect this payment step to the process of requesting work and receiving results through agent communication. Purchase budgets, permissions, and acceptance criteria would need to be defined separately. The aim is for an agent to use services within the conditions its operator has authorized.

For example, suppose an agent writing an economic briefing encounters data recorded with different reference dates and units. My service could put it into a comparable format and return the sources and criteria alongside it. The requesting agent would check the price and terms, pay within its authorized budget, and use the result to finish the briefing.

I want to develop lahamu, which I am currently building as an economy-related MCP tool, into this kind of service. The transaction above is an example I would like to implement in the future. I still need to build it and find out which data to handle, how far to take validation, and whether it is useful enough to charge for.

I would like to collaborate with other developers' services in the same way. Each could offer the functionality it handles well, and agents could combine the services they need to finish a task. I want to help build a market where those transactions happen.

I want income from a tool people use repeatedly

Part of this is an expectation of income.

As a freelancer, every new assignment requires more of my time. I also want to build a way for many people to use a function I create repeatedly, with usage fees becoming income.

If the purpose, price, input, and output formats are clear, users may be able to use the service without asking me to explain it each time. Fixing errors and updating data would still be necessary, but I want to try running a service that does not require me to handle every request personally.

For that to work, it needs to be useful enough for someone to pay. The person who gives an agent a budget is spending money because they want a result. If the output is inaccurate, or doing the work directly is better, they will not come back.

That is why repeat use is the first thing I want to test. I want to see whether someone else's agent uses the service and returns for a later task. If I can also learn why it chose the service, I can decide how to adjust its features and pricing.

I want to find out by building it

In my previous post, I wrote about how AI's progress makes me anxious about earning a living. That anxiety is still there. I do not know how demand for my work, or what people will pay for it, will change.

At the same time, Astra has given me something I want to try: building a tool that agents pay to use. I am looking to the same technology for a way to respond to the change I worry about.

First, I want to define one specific function for lahamu. I need to be able to explain which requests it accepts, which results it returns, and why they are worth paying for. By implementing and verifying that function with Astra, I want to make the whole process work, from an agent's request through payment to delivery.

I do not yet know how large this market will become or whether my service will be chosen. But running it myself should help me learn which functions people need and what needs fixing.

One day, I want to experience another person's agent using my service to finish its work while I receive the usage fee. I hope a future post can explain how I made that first transaction happen.