The diff-first approach: agentic AI for cross-type transition training

Operators often face challenges when transitioning between systems in industry, requiring knowledge and skill transfer. Addressing this issue involves multiple factors, including personnel, organisation, knowledge, and technology, often resulting in lengthy and inefficient processes. Inspired by transfer theory, we define the knowledge resources and cognitive support needed to solve the cross-type transition training (CTTT) problem from the perspective of differences. Agentic AI's planning, execution, and multimodal generation capabilities offer a promising solution. We introduce the Diff-First Approach, which leverages cross-type differences to build a knowledge transfer framework. We propose a multi-level multimodal knowledge resource structure for retrieval, with the Diff-First Thought Chain guiding AI agents’ reasoning and providing enhanced multimodal cognitive support to users. The aircraft type transition process in aviation is used as a case study, and DELTA-T is proposed as its implementation. This system utilises agents to perform transfer target identification, knowledge retrieval, and multimodal cognitive support generation. The agentic AI framework enhances recognition and multimodal retrieval in knowledge transfer tasks, achieving 92.7% accuracy in image editing and favourable subjective ratings for over 93% of generated content. These findings demonstrate that the proposed method effectively supports CTTT tasks in specialised knowledge domains, highlighting a promising human-AI training paradigm.