The more AI amazes me, the more I worry about making a living

Using AI these days leaves me both amazed and uneasy. I am glad I can finish work faster, but I also think that clients could now do some of the work they used to give me themselves.

GPT-6 Astra has made both feelings stronger. I can ask it, in a conversation, to write code, compare sources, and check results. These are tasks I used to handle myself across several windows.

I also updated this blog with AI. We examined its structure, changed the code, and checked the build and deployment results. Some things needed fixing again, including the sidebar's clickable areas and the theme button. Reviewing the results was still necessary, but I was able to take the changes through to deployment on my own.

Having access to tools like this is exciting as a developer. At the same time, I worry about how much they will change the way I earn a living from software development.

When AI can handle more of the work

OpenAI describes Astra as a model for complex tasks such as reasoning, coding, computer use, research, and document writing. Many of these are things I do every day. Official GPT-6 Astra documentation

It has become less daunting to research unfamiliar subjects and turn ideas into working software. That is welcome news for a developer working alone. But the people who used to hire me may also be able to get what they need with the same tools, or complete projects with smaller teams.

I do not know when AGI will arrive. Even before then, the number of people needed for development, its cost, and the way we work could change. For me, the more immediate questions are whether I can find my next contract and whether it will pay as well as the current one.

Is it enough to rely on income from work?

The money that supports my life still comes from doing work and getting paid for it. I have expected that gaining experience and taking on more difficult work would make life more stable. Now I also think about how quickly AI could learn to handle the work I know well.

If I can produce the same result in less time, I might be able to take on more work. On the other hand, clients might cut their development budgets and team sizes. Higher productivity does not guarantee a higher income for me.

The second possibility frightens me. I worry that more people may struggle to find enough work even when they try to adapt.

That does not mean there is evidence that most people will lose their jobs. The International Labour Organization's 2025 study does not equate occupations exposed to generative AI with occupations that will disappear. It emphasizes the likelihood that the tasks within many jobs will change. ILO, Generative AI and jobs: A 2025 update

I cannot predict these changes precisely, but I want to develop more ways to earn money. That is why I am continuing my current work while preparing income sources that do not depend entirely on my working hours.

Looking for something I can sell repeatedly

Lately, I have been exploring automated income using LLMs. I want to build something that many people can use after I create it once, with orders, payments, and delivery handled automatically.

I have not established that income yet. I am researching and building different things to decide where to spend more of my time.

Even when sales are automated, there is still a service to run. Incorrect results need fixing, outages need resolving, and API and server costs need managing. Users also need a reason to keep paying. Having AI help with development does not create paid demand by itself.

Still, I want to try. While I worry that AI may reduce demand for development work, I am also looking for ways to use the same tools to build and run products on my own.

lahamu: Economic tools for agents

The lahamu project I am building is an economy-related MCP tool for AI agents. I want to provide functions that agents need when working on economic tasks.

The goal is to charge for usage. I am considering a model in which agents pay USDC through the x402 protocol when they use lahamu. That does not mean I am already earning money this way.

x402 uses HTTP's 402 Payment Required response to communicate payment requirements for APIs and web resources, allowing clients to pay programmatically under those terms. Agents can use it to call a service and pay its usage fee. Official x402 documentation

I want to run a service where a tool I have built does the work without my having to handle every request, and its usage fees become income. First, I need to find out which agents need the functionality and whether it helps them enough to justify paying. What matters is whether they use the tool repeatedly after the first integration.

I wrote more about connecting agent collaboration and payments in the agent market I want to build with Astra.

Why I am preparing a blog and an Inflearn course

Even when I solve a problem at work, it is hard to explain the process on the next project if I have not recorded it. Writing down where my judgment was wrong and how I decided what to change could help someone facing a similar problem. I want to use those records to explain my experience and develop it into courses or products.

That is also why I reorganized this blog. I want to write about things I have experienced and checked myself, and record what happened when I tried something, including why it failed.

I am also preparing a course for Inflearn. I want to combine my experience as a full-stack lead instructor and mentor with my experience developing services alongside AI. Its working title is “Shipping a Production Full-Stack App with AI Agents.”

It is aimed at developers who have used React, APIs, and Git but have not deployed a service on their own. I plan to cover requirements, verification of the implementation, testing, and deployment. I have outlined the curriculum and started the first lesson's materials. I describe the practice project and production process in my course development notes.

I have concerns here too. As AI tools change, explanations can quickly become outdated. Along with specific buttons and commands, I want to teach the process of deciding what to build, checking the results, and finding why something failed. That is where I can use the experience I have gained in professional development and education.

I am considering YouTube and TikTok as well, but for now I will start with the blog and the course. If I launch several channels at once, I suspect I will struggle to make enough progress on any of them.

Preparing while continuing freelance work

Professional development still requires understanding requirements and existing systems, coordinating different requests, and checking that the result works in actual business operations. My experience still helps me do those things.

But I cannot guarantee that this experience alone will keep my income stable. As AI becomes capable of more work, I will need to change the way I work too.

I intend to do my current work well while preparing articles and a course, and looking for features in my personal projects that I can sell repeatedly. I do not expect these efforts to replace my freelance income immediately. I want to try them and spend more time on the ones that show promise.

I welcome AI's progress and am curious about what I will be able to build. I also worry about my next contract and my income. Right now, both feelings are with me.

I cannot yet write that I have found a way to earn automated income. Instead, I will keep recording why I started preparing, what I am building, and what results I get.