[AI One-Bite News] 'AI Agent' Resembling 'Javis'... Differences from LLM

Autonomy in Work Planning... AI Collaboration
Insurance Claim Review 2 Weeks → 3 Days... Efficiency Up with Automation

There is a term that has been appearing as frequently as large language models (LLMs) recently. It is the 'Artificial Intelligence (AI) Agent.' It is also referred to as 'Autonomous AI.' Unlike LLMs that provide answers to relatively simple questions, AI agents can independently perform complex tasks.

AI agents can autonomously work to achieve designated goals. When a specific task is requested, they gather the necessary data, plan the methods and sequence of the tasks, and execute them. They may repeat this process until the objective is achieved.

For example, let's say you ask an AI agent, "Make a dental appointment for me on Friday after 4 PM or any free time on Saturday morning." The AI agent starts by opening the user's calendar to check available time slots. Then, it accesses the dental appointment system to book the appointment at the selected time and sends a message to the user saying, "I have booked a dental appointment for 10 AM on Saturday."

The artificial intelligence named Jarvis appearing in the movie Iron Man

To perform various tasks such as schedule checking, booking, and message sending, AI agents utilize user data and integrate with external systems. Tasks can be divided among different agents. For example, a schedule-checking agent, a booking agent, and a message-sending agent collaborate by exchanging information. In this process, AI agents may use LLMs suited to each function.

Since human intervention is reduced, work efficiency can improve. For instance, while out of the office, an AI agent could attend meetings on your behalf and summarize the necessary information, or while driving, you could listen to emails read aloud by the AI agent and give voice commands to handle tasks. This is like the realization of 'Jarvis,' the reliable assistant from the movie Iron Man. If AI agents are linked with robots, they could even perform physical tasks.

Samsung SDS, which is preparing a 'Personal Agent' feature for its generative AI service 'BriTy Copilot,' cited insurance work as an example. Typically, when a car accident occurs, the process from accident reporting to insurance payment involves 22 steps and at least nine customer communications. If agents specializing in accident investigation, vehicle repair, and accident compensation divide the tasks, the process that used to take up to two weeks could be shortened to within three days.

Changseong Jung, Head of Samsung SDS IW Business Team (Executive Director), explained, "If Copilot was a diligent assistant that understood and responded to queries requested in natural language, the Personal Agent will be a smart and reliable helper that understands the context and patterns of work and proactively and actively assists with tasks."

Domestic and international companies are competing beyond LLMs with AI agents. According to global market research firm Grand View Research, the global AI agent market size is expected to grow at an annual rate of 42.8%, reaching $70.53 billion (approximately 93 trillion KRW) by 2030.

There are challenges to overcome. First, becoming a multiplayer itself is a complex task. Multiple LLMs must be used and seamlessly integrated with various software (SW) and external application programming interfaces (APIs). This process requires significant computing resources. Since diverse data must be utilized according to the situation, privacy and security issues may arise. An industry insider predicted, "Until now, there has been little difference among LLM-based chatbots or Copilot solutions, but with AI agents, meaningful gaps may emerge depending on data utilization and system integration."