In many organizations, essential operational knowledge is not formally documented but passed on through everyday human interaction, including informal dialogues, spontaneous problem-solving, and one-to-one guidance. As experienced employees retire, change roles, or leave, this tacit knowledge often disappears. This poses a potentially fundamental threat, or at least a serious challenge, to organizational continuity. This paper presents the Learning Cycle, an early-stage AI-driven framework designed to capture and structure workplace knowledge. The framework consists of two components: knowledge intake and knowledge delivery. In this paper, we focus on the intake side, examining how informal knowledge is captured, processed, and prepared for integration into a learning management system (LMS). Using speech recognition, large language models (LLMs), and semantic clustering, the system processes various sources, including recordings of authentic workplace dialogues from guided training scenarios. Examples include mentoring moments, troubleshooting discussions, and task explanations for new employees. While these conversational interactions serve as the primary data type in our case study, the system is designed to handle a broader range of multimodal inputs, including videos, documents and images. Our approach builds on work in organizational learning and knowledge management and extends it using AI to process informal, multimodal knowledge sources. We illustrate how selected segments of workplace dialogues were processed by the system and subsequently examined by us to understand how the AI transformed them into LMS-ready content. Rather than focusing on system architecture, we highlight intermediate representations and design decisions that shape the resulting content.