B, the chief technology officer (CTO) at Korean company A, used 10 million won ($7,200) worth of tokens alone over the past month. The entire team under B consumed a total of 60 million won ($43,000) worth of tokens during the same period. Token costs ballooned as employees developed various artificial intelligence (AI) agents and ran them around the clock.
As AI transformation (AX) has become an essential survival strategy for companies, Korean firms are now paying attention to tokens, the basic unit of AI computation. While they encourage more employees to use tokens and apply AI to their work, concerns are also growing over the resulting token costs. Industry watchers say efficient token management is rapidly emerging as a critical element of corporate AI governance.
According to industry sources on Wednesday, Korean companies such as Upstage and Wrtn are focused on getting employees to actively use AI in their work. The Korean market is still in the AX push stage, and companies believe spreading AI experience is more important than immediate token management. "There are even stories here of someone using 100 million won worth of tokens," an AI industry official said. "We are looking at the proportion of employees who maximize token usage, but the numbers are still not high."
However, a growing number of companies are proactively seeking token management measures. LG CNS recently developed and began operating a dashboard that shows employees' token usage. The system receives token usage data from OpenAI as of two days prior, then reprocesses and displays it based on its own criteria. In addition to the dashboard, LG CNS is working on guidelines to help employees use tokens efficiently and boost work productivity.
Real estate data tech company Big Value went a step further and developed a token management AI agent. Big Value last month launched "Bokdeokbang Gajae," a service in which AI agents, each assigned a specific role, converse with one another to derive answers when customers seek real estate consultations. As conversations among AI agents increased during actual operations, token usage surged. In response, the company appointed an AI administrative agent to manage token costs. The administrative agent tallies the previous day's token usage each day and prepares cost reports.
AI startup Liner issues light large language model (LLM) API keys to developers so they can use AI. This approach allows developers to use tokens regardless of the LLM type. The measure is designed to prevent service development delays caused by token shortages in specific models.
Companies are paying attention to token management because token usage has surged as even non-coding employees now build AI agents and run them 24/7, 365 days a year. On top of that, Anthropic, OpenAI and Google have successively introduced pricing schemes based on token usage, increasing the cost burden. Token optimization strategy has become a variable that could determine the success or failure of AX. "We allocate tokens per team, but if a particular employee consumes a lot of tokens on low-productivity tasks, highly productive employees end up facing token shortages," an AI startup official said. "We haven't placed any particular restrictions yet, but concerns over token management are bound to grow."
Choi Kyung-jin, a law professor at Gachon University and president of the Korea Artificial Intelligence Law Association, said, "Developers who already have a high understanding of AI and use it well tend to overuse tokens. Rather than developers, it is necessary to distinguish and give employees who have not used AI much the opportunity to use it extensively and go through trial and error."
