Discussion of artificial intelligence in education often moves too quickly from access to tools to claims of improved efficiency. The interview evidence in this study does not support such a direct conclusion. This article explores how teachers and administrators in Shandong, China, described their use of AI in technical and vocational education and training (TVET). It focuses on the situations in which AI appeared to reduce routine work, and the situations in which it generated additional checking, explanation, or risk. Eight semi-structured interviews were conducted across secondary, higher, and adult vocational institutions. The interviews were recorded and transcribed in Chinese. English working versions were then prepared with AI assistance for analysis and reporting. During coding and manuscript preparation, the first author returned to the Chinese transcripts to verify key terms and selected quotations. The data were analysed in NVivo through iterative coding, memo writing, and comparison across teacher and administrator accounts. Findings are organised through a Constraints–Mechanisms–Enablers–Outcomes (CMEO) model. Participants did report some efficiency gains, but only in a limited sense. These gains were mainly associated with less time spent on first-draft writing, formatting, and routine screening. They also described new work. Teachers still had to verify technical accuracy, explain AI-supported feedback to students, and adapt outputs to local equipment, syllabi, and safety rules. In higher-risk settings, AI use was accepted only when teachers retained decision authority and could justify the result. For that reason, this paper treats AI-enabled teaching efficiency as a situated professional judgement rather than as a direct measure of productivity.