Evolution in the Field
Henry Kim ·
AI is evolving at an astonishing pace even at this very moment. Ask it any question, and it answers with more expertise than anyone else in the world. However, the reality experienced firsthand in the field is different. From my perspective, working on the front lines, AI has not yet achieved complete evolution in practical application.
The first limitation is that live, on-the-ground data is not fully conveyed to AI. In work, there are subtle contexts that are difficult to explain verbally. There's an abundance of tacit knowledge that people take for granted and therefore don't bother to record. It is practically impossible to re-interpret, refine, and spoon-feed all this information to AI.
The second is the structural limitation of current LLM models. API-based LLMs are like a partner whose memory resets every day. Each time, we have to inject our context and data from scratch. While we try to bridge this gap with vector data technology and prompt engineering, there are clear limitations. There's a lack of an organic process that immediately reflects real-world problems and learns autonomously. Separately training models yields results too slowly, significantly reducing their utility in a business environment where speed is critical.
The third issue lies in the implementation phase. We still lack sufficient experience in successfully integrating such AI systems across entire businesses. IT developers are still confined to traditional code logic with clear causal relationships, and business professionals don't yet grasp how to combine their rich experience with AI technology. To bridge this gap, metacognitive ability—the awareness of what one knows—is absolutely essential. The key to successfully applying AI in the field is not the technology itself, but human insight: asking the right questions and providing clear direction.
These limitations are the stark realities I face daily in the field. For now, I'm finding workarounds, personally intervening and making continuous fine-tuning to bridge the gaps. I cannot definitively say how quickly this transitional period will end. However, in the not-too-distant future, companies will adopt business AI as if hiring new employees. Simply by connecting them to on-the-ground data, they will evolve into AI perfectly optimized for our organization within days. This is an undeniable future.