[Market Pulse] AI's Second Act Is Transforming the Future of SMEs

From Agentic AI to Physical AI
Ontology Will Shape Corporate Competitiveness

Park Youngsun, Chair of the Strategic Economic Advisory Committee and former Minister of SMEs and Startups

Artificial intelligence (AI) is entering a new phase. After the era of generative AI, which answered questions and wrote documents, AI is now evolving into agentic AI, which understands goals, makes plans, and carries out tasks. The next step is physical AI, which perceives the real world and operates robots and machines. The AI revolution, which began in digital spaces, is now moving into manufacturing sites, logistics, and the heart of industry. In the face of this change, small and medium-sized enterprises need to ask: "In an industry transformed by AI, what should our company change first, and what should we connect to AI?"

First, companies should change how they work by using agentic AI. Small and medium-sized enterprises do not need to build massive AI systems from the outset. They can start with repetitive, time-consuming tasks such as preparing quotes, responding to customer inquiries, checking inventory, managing delivery schedules, and writing quality reports. For example, they can connect AI to a workflow that checks inventory when an order comes in, reviews the production schedule, and then proposes a delivery date to the customer.

The important thing is not to stop at using AI as a simple document-writing tool, but to connect it to workflows from beginning to end and have it support repetitive tasks that people previously handled every time. People should retain final authority over major decisions and exceptional situations, while companies measure how much they reduce processing time, costs, and errors, and expand the scope of AI's application.

AI Must Expand Beyond Simple Document-Writing Tools to Connect and Support Work

Second, organizations must also change to coexist with AI. AI transformation is not just an IT department's responsibility. If data and workflows are disconnected across sales, procurement, production, and logistics departments, the benefits of AI will inevitably be limited. Connecting the entire process from order to delivery makes it possible to predict inventory shortages and production disruptions and respond proactively.

Small and medium-sized enterprises can start by forming a small AI transformation team that includes the business owner, frontline employees, and IT staff. Employees who know the workplace best should be at the center of AI adoption. The goal is not to replace people with AI, but to reshape work so people can move away from repetitive tasks and focus on improving quality, serving customers, and identifying new business opportunities.

AI-generated reference image

Third, more important than the quantity of data is the ontology that connects its meaning. When it comes to AI performance, the accuracy of a company's data and whether it can be connected are also important. Companies should identify data scattered across Excel files, accounting systems, enterprise resource planning (ERP) systems, production equipment, and quality inspection records, and standardize basic identifiers such as product codes, customer names, and equipment numbers.

In particular, defect rates, equipment downtime, operating conditions, and repair histories accumulated over many years by small and medium-sized manufacturers are competitive assets that are not easily available elsewhere. If the experience and know-how of frontline employees are captured as data, the organization as a whole can reuse problem-solving methods that once depended on the memory of individual employees. Access permissions and data security should also be designed from the outset. Simply collecting a large amount of data is not enough. If the sales, production, and procurement departments each manage the same product using inconsistent data, AI may struggle to recognize that the data refers to the same thing. Simply bringing together data recorded in different terms across departments is not enough to create AI that understands the company as a whole.

A Knowledge Map of Concepts, Entities, and Their Relationships Can Improve AI's Ability to Perform Tasks

This is where ontology comes in. An ontology is a knowledge structure that defines the concepts and entities a company uses, as well as the relationships between them. Put simply, it is a "digital map" that enables AI to understand a company's work and operations. It systematically links products, components, equipment, workers, suppliers, and customers, defining what they are and how they relate to one another.

For example, suppose a piece of equipment on a production line stops working. If the data is disconnected, AI can only check the equipment downtime record. But the situation changes when an ontology connects the equipment to the production process, components used, suppliers, past breakdowns, and delivery schedules. AI can trace the cause of the failure, determine which orders could be affected by production disruptions, and assess options such as replacement components and delivery schedule adjustments. Data is transformed from a simple record into knowledge that can be used for actual decision-making.

This is the foundation for improving agentic AI's ability to perform tasks and supporting physical AI's decision-making on the factory floor. An ontology does not mean that a company must first build a massive data platform. It can start with small steps, such as standardizing product codes, equipment numbers, and defect types, and defining how workplace data relates to specific workflows. This process turns a company's experience and know-how into assets that the organization can reuse.

Understanding and Problem-Solving Matter More Than the Scale of AI

Fourth, physical AI represents the next opportunity for small and medium-sized enterprises. While conventional automation performs repetitive tasks according to fixed rules, physical AI uses cameras and sensors to perceive its surroundings and adjust its actions to the situation. It can identify the location of a component so a robot can pick it up, detect surface defects in products, and analyze changes in equipment vibration and temperature to predict potential failures.

There is no need to convert every process to AI. Companies can first validate its effectiveness in processes where labor shortages are severe, defect costs are high, tasks are highly repetitive, or workers face significant risks. Safety measures, including worker detection and emergency stops, are also essential. Competitiveness in the AI era does not depend solely on who has the largest AI system. What matters is who understands their industry and operations most deeply, gives meaning to data, and uses AI to solve problems more effectively. Companies that change how they work, connect data through ontologies, and transform their operations with physical AI will gain a new competitive edge in AI's next act.

Park Youngsun, Chair of the Strategic Economic Advisory Committee and former Minister of SMEs and Startups

This content was produced with the assistance of AI translation services.