AI Agents and the Transformation of the Financial Industry
Insight | 2025-04-18
16 minute read
Over the past two years since the emergence of large language models (LLMs) as the foundation for generative AI, the pace of progress has accelerated rapidly. In particular, AI agents capable of autonomously executing tasks based on generative AI have garnered increasing attention. These autonomous AI agents integrate generative AI with traditional AI technologies, offering flexible and advanced capabilities. As a result, AI agents can now handle complex tasks and meet a wide range of user needs.
On the other hand, the financial sector has a strong affinity with AI in both tasks that require language processing and tasks that involve data processing, such as analyzing vast amounts of transaction data. Currently, the financial services industry, which continues to undergo transformation through AI, is entering an era of change defined not just by generative AI, but also by AI agents.
This article will explore the structure and functionality of AI agents, examining their impact on the financial industry and how human-AI agent collaboration is shaping the future of financial services.
1．What are AI agents?
There are various definitions of AI agents, but simply put, an AI agent is a digital system capable of interacting independently within a dynamic environment. While such systems existed even before the advent of large language models (LLMs), AI agent systems based on generative AI (LLM-based) can understand given goals and contexts, create their own action plans, and use online tools and data to complete tasks. Furthermore, these agents can collaborate with other agents and humans, learn, and improve their performance through memory. In other words, AI agents consist of not only LLMs but also components such as database access, API integration, rule-based engines (traditional AI), and interfaces with external systems(1).
Figure 1 illustrates a conceptual diagram showing the differences in workflows between generative AI (LLM) and AI agents. With generative AI, humans are typically involved in the workflow, directly influencing the AI system's decisions and tasks. On the other hand, AI agents can autonomously complete the tasks assigned to them, with humans serving a supervisory role, intervening only when anomalies or unexpected situations arise. However, it’s important to understand that inference models like OpenAI-o1 can autonomously perform tasks that require multiple steps from a single prompt, making a direct comparison with AI agents difficult and highlighting that this is an evolutionary process.
