Keeping employee-created AI agents under company control

Source: Seoul Economic Daily — Genesis Lab "Targeting Japan with AI Agent OS Free of Security Concerns"

Lee Young-bok, CEO of Genesis Lab, met with Seoul Economic Daily to announce the business plan for Agentria, an enterprise AI agent operating system (OS).

With the spread of vibe coding, even non-developers are creating task automation agents using Claude Code or GPT Codex. The first concern for enterprises here is not performance, but internal information. The question standing in the way of adoption is whether the agents created by employees will leak sensitive data outside. Agentria is a product that started from solving that very concern.

When data and manuals scattered across the company are gathered into a folder called Slipbox, the company's own unique ontology is created. When an employee attaches a Large Language Model (LLM) to this, it becomes an agent running within the company's governance. The process the agent went through to produce results is displayed using a black-box function. Employees can immediately find the step where an error occurred.

About 20 companies, including electronic component manufacturers and financial companies, have entered the Proof of Concept (PoC) stage. CEO Lee shared that clients are first applying it to tasks such as reviewing individual company shareholders and financial statements, as well as evaluating recruitment documents.

The product targets large enterprises, public institutions, and military units. They determined that organizations more cautious about adopting external AI models are the ones looking for products with control mechanisms in place. Genesis Lab is also preparing to enter the Japanese market with this product. They are discussing PoCs with local AI and retail tech companies and plan to establish a Tokyo subsidiary within the year.

CEO Lee said, "To succeed, we must successfully implement even the tacit knowledge of Korea's 'work experts' (il-jal-leo) into AI agents."