Organizational Design Guide for Adopting Agentic AI - LG AI Research BLOG

The AI paradigm is rapidly shifting from Passive AI, which simply provides answers, to Agentic AI, which sets its own goals and uses tools. Corporate interest is also expanding beyond simple LLMs toward intelligent agents that understand business Goals and execute Tools and Workflows.

However, as agent autonomy increases, the problem of opacity in the decision-making process arises. It becomes difficult to trace why an agent made a particular decision, where errors occurred during execution, and whether policies and regulations were complied with. To solve this, a logical structure such as an Ontology must be layered on top of the probabilistic generative model, the LLM, to control agent behavior and make it observable.

Research that drew attention at AAAI 2026 is focused on Structural Reasoning, which combines Knowledge Graphs (KG) with LLMs to improve reasoning performance and optimize cost. Representative approaches include the following.

• PathMind: Reduces API costs and increases efficiency through a 'Retrieve-Prioritize-Reason' framework that selects the reasoning path most likely to reach the correct answer.
• DoM: Improves the reliability of LLM answers by compensating for the limitations of incomplete knowledge graphs through a multi-agent debate mechanism.