[Sung-hyun Cho, head of technology at Databricks Korea] South Korea is one of the markets with the steepest pace of AI adoption worldwide. According to the Ministry of Science and ICT, South Korea ranked first among OECD members with a 28 percent AI technology adoption rate, and also led in the adoption of big data analytics and the Internet of Things.

OpenAI has also said South Korea is a market with a large number of paid ChatGPT subscribers, second only to the United States. AI discussions among South Korean companies are now shifting beyond simple chatbot adoption to how to combine and operate AI models tailored to industry-specific data and business context.

In the past, organisations were at the stage of experimenting with individual chatbots on the periphery. Now, multi-agent systems in which multiple AI agents collaborate to design, execute and scale digital workflows are emerging as a new operating model.

The key point is not to apply the most expensive and highest-performing single model to every task. What companies need is not the latest model, but the capability to select the "most suitable and reliable" model that can accurately convey their business domain knowledge.

According to the "AI Agent State of the Union" report published by Databricks after analysing anonymised data from 20,000 companies worldwide, use of multi-agent systems has increased 327 percent in recent months. It also said 78 percent of companies use two or more large language model families at the same time, including GPT, Claude, Llama, Gemini and Qwen. This shows companies are building flexible AI stacks by combining the strengths of various models for different use cases rather than relying on a single model.

Automation creates scalability

Leading companies worldwide are already designing structures in which specialised agents set their own plans and autonomously execute complex tasks. Such systems break down large goals, assign sub-tasks to domain-specific agents, and integrate structured and unstructured data to continuously improve outputs.

In South Korea, AI and data platforms have already taken hold in core business processes. For example, fashion platform Musinsa used the Databricks Lakehouse to provide personalised recommendations based on vast consumer data and strengthened a data-driven decision-making environment.

What stands out is that the spread of these AI agents is driving fundamental change in companies' underlying data infrastructure. AI agents have evolved beyond simply consuming data to become active participants in database operations. According to the report, 80 percent of databases worldwide are currently built by AI agents, and the share reaches 97 percent in testing and development environments.

In this way, AI agent-based automation goes beyond reducing simple repetitive tasks and helps companies build scalable operating systems by connecting more data and workflows. Ultimately, the value of automation does not stop at improving efficiency in individual tasks, but in laying the foundation to quickly improve organisation-wide decision-making and customer experience.

Governance is a prerequisite for AI expansion

The faster AI use expands, the more important "evaluation" and "governance" become. To safely move AI from the experimental stage into a real operating environment, it is not enough for the model itself to perform well. A company-wide control system that can ensure reliability, transparency and accountability must be a prerequisite.

According to the Databricks "AI Agent State of the Union" report, companies with a framework to systematically evaluate model outputs had a deployment success rate in real operating environments about 6 times higher than those without one. The report also found companies that invest in strong AI governance were 12 times more likely to successfully move AI projects from the experimental stage into real operating environments. The figures show AI governance is not regulation that holds innovation back, but a powerful catalyst that enables AI to spread safely.

Only on such a thorough governance foundation can companies realise the true value brought by "model flexibility". Some models are strong in high-level reasoning, while others perform better in multilingual content generation, code development or industry-specific summarisation tasks. By using multiple model families within an integrated architecture without being locked into a single vendor, companies can simultaneously optimise performance, cost efficiency, regulatory compliance and business resilience.

South Korean companies' AI shift has now moved beyond the stage of adopting a single chatbot or model, and entered an advanced process of flexibly combining models and agents to fit industry data and business context. The core competitiveness that will determine future business continuity also depends on how systematically these diversified elements are orchestrated.

Companies seeking a successful AI shift need a broad perspective that embraces model diversity as a powerful "strategic asset" rather than complexity to be avoided. The immediate task is to overhaul fragmented data infrastructure and build evaluation and governance in a balanced way from the outset on a reliable data platform. When multi-agent systems are established as the core foundation of a digital operating model beyond a temporary technology experiment, AI will take root as scalable power that innovates company-wide decision-making and customer experience beyond a tool for simple automation.