Something I heard recently from a graduate of a software-focused vocational high school stayed with me. Many students, he said, do not have proper access to paid frontier models. Some do their assignments with Claude and Codex subscriptions paid for by their parents. Others wrestle with the same assignments within the tight limits of free models. Same school, same desks, same Wi-Fi. The difference is which tool is running beside them. Students who entered the same school to become developers are starting from different lines.
This divide is not confined to that classroom. It is happening at the same time in university lecture halls, bootcamps in Jakarta, vocational schools in Lagos, and on the laptop of someone trying to change careers at forty. One side tackles assignments with a search bar and a free model. They copy error messages and pause YouTube tutorials to type along. The other side opens a frontier model. These tools are no longer teaching assistants that just explain things. They read repositories, edit files, run tests, review failure logs, and make more changes. Give them an issue, and they come back with a PR. Both sides have the same assignment, but one works alone while the other starts with a few colleagues sitting beside them.
The productivity gap created by agentic tools is stark. Prototypes that used to take weeks are built in hours. People get codebases they could never have tackled alone up and running over a weekend. Doubling or tripling productivity has become a common story among working developers. For beginners, the multiple is even larger. It is the difference between zero and one. With AI, they can get something working that they could not even have started alone.
This is no longer confined to computer science departments. Marketers code data dashboards. Lawyers write scripts to review contracts. Doctors run their own patient data analysis tools. Designers produce interactive prototypes in code. Coding is becoming a form of literacy rather than a profession. The price of an AI coding tool is therefore not the price of a developer's assistant. It is the price of a new writing tool for the 21st century.
Until last year, there was still a belief that with a laptop and an internet connection, you could somehow learn. Gaps in English, equipment, and mentorship had always existed, but the path to building skills by digging through public knowledge was open. That was what made it possible for a student in rural India to take MIT lectures and get a job in Silicon Valley.
AI shakes that belief. Learning used to be closer to learning a language. You learned variables, conditionals, and functions, then wrote your own sentences. Now an interpreter and collaborator sits beside a student who has not yet learned all the grammar. The student does not start with an empty file. They start with something already running. The hardest part for a beginner is not the grammar. It is not knowing what they do not know. A good AI tool gets them through that uncertainty. Who can try more things in the same amount of time, fail faster, and fix things sooner? This is where the gap opens up.
That makes the price issue more acute. And one thing needs to be clear. It is hard to do real development on a $20-a-month entry-level subscription. When an agent reads an entire repository, edits several files at once, and runs tests repeatedly, tokens run out fast. An entry-level subscription gives you enough for a few questions to a chatbot, but nowhere near enough to work on a real project for days. The tools used by people actually making things are higher-tier plans like Claude Max or ChatGPT Pro, at $100–200 a month. Add the cost of major API calls, and you can easily spend 300,000 won a month. For a student, that is more than an entire month's allowance.
And the price is the same everywhere in the world. At $200 a month, an American student might stretch their budget to try it once. For a Korean university student, it is part of the rent. For a young Indonesian, it is close to a month's wages. For many young people in sub-Saharan Africa, it is several months' wages. In quite a few places, the tool that helps you work better costs more than you earn from working. Even two people with the same laptop and the same internet connection start from different lines.
This is where we need to revisit the discussion of basic income. Basic income asks how to share what society has already produced. It is a question of distribution. But what is happening now comes one step before that. The ability to make something at all is being divided by price. Someone with a Max-level tool beside them can go from idea to code, code to product, and product to revenue in a matter of days. Someone without the tool spends a month with the same idea still in their head. Before distribution even begins, a gap opens between those who can create the pie to be shared and those who cannot.
That is why basic AI comes before basic income, not after it. Before society creates something to share, we should guarantee a minimum starting point for the ability to create. Every citizen should have at least one colleague who can help take their work a step further. This is not welfare. It is infrastructure. Just as compulsory education was in the 19th century, and electricity and running water were in the 20th.
But here the rules change once more. Handing everyone an entry-level subscription does little. The people actually making things are using Max-tier tools. So the real policy question becomes narrower. Do we give every citizen the same shallow tool, or find a way to guarantee Max-level tools for people who are actually producing? Students, founders, researchers, solo business owners, and workers trying to change careers are putting time into making something. They need a higher tier, not an entry-level subscription. Whom do we define as a producer, and in what quantities do we supply them with tokens? This will be one of the central questions of industrial policy over the next decade.
Unlike Spotify or Netflix, AI is hard to solve through country-specific pricing. Once music and video have been made, an additional viewing or listen costs almost nothing. AI is different. Every time a model answers, there is a real token cost in GPU time and electricity. The same question costs the same whether the user is in Korea or Bangladesh. That is why a single global price is hard to break.
Basic AI, then, is not something that can wait for international organizations to reach an agreement. It is a strategy individual countries must take on themselves. The question is who will pay the token costs for their own citizens. A government might buy AI access in bulk and distribute it to citizens, sign a national licensing agreement with a model provider, or run open-source models on domestic infrastructure. In the end, it comes down to a country's finances and resolve. With open-source agent infrastructure layered on top, the underlying token costs can be controlled domestically. And this AI should be provided as a shared working environment, not a question-and-answer tool.
One country has already moved. In 2025, the UAE reached an agreement with OpenAI to provide ChatGPT Plus free to all citizens and residents. What matters is that an oil-producing country chose to share its oil revenues with citizens through AI access rather than cash dividends. It chose the minimum conditions for productive capacity before distribution.
Korea is now discussing how to spend the extra tax revenue brought in by the semiconductor supercycle. There is talk of distributing a certain amount per person, of another round of livelihood relief payments. Instead of a dividend that disappears after a single round of spending, the money could guarantee a Max-level AI colleague for citizens who are making things. Providing entry-level tools to all 50 million citizens for a year would cost about 1.6 trillion won. Guaranteeing Max-level tools to one million people for a year would cost about 2.4 trillion won. Either way, for a fraction of the cost of one-time cash support, society would be left with a fundamentally different foundation for production.
Just as the UAE chose the beginning of the AI era at the end of the oil era, Korea must choose whether to let the gains of the semiconductor cycle pass as a one-time consolation payment or leave them as infrastructure that raises every citizen's starting point. The question is clear. It is not whom we give money to, but how far we extend access to something that works alongside people, and how much of its capacity we give them. Before basic income, basic AI. We have to be able to make things before there is anything to share.
