The Token Economy

Artificial intelligence is neither free nor unlimited. Every time a model reads an instruction, writes a text, generates code, or makes a decision, it consumes small units of information known as tokens. Although tokens are usually treated as a technical detail, this paper presents them as an important new economic variable.

The main purpose of the study is to examine how the cost of using artificial intelligence affects business decisions. More specifically, it explores when automating a task is economically worthwhile and when employing a person remains the better option.

One of the paper’s central findings is that lower token prices do not necessarily make automation cheaper. Advanced AI systems can plan, repeat processes, correct mistakes, and use external tools. As a result, they may consume hundreds or even thousands of times more tokens than a simple request. In some cases, an automated task could therefore become more expensive than the human worker it was intended to replace.

The paper also highlights an apparent contradiction. While the cost of processing information has fallen rapidly, the total use of AI and investment in data centers continue to grow. The reason is that when a technology becomes cheaper and more accessible, it begins to be used across a much wider range of activities.

The conclusion is that the future of work will depend not only on what artificial intelligence can do, but also on how much it costs to use. In the near term, many companies are likely to adopt forms of cooperation between humans and AI rather than fully replacing their employees. In this new environment, managing computing resources efficiently may become just as important as having access to the technology itself.

Read the full paper on SSRN.