本研究之貢獻在於將認知師徒制三法轉譯為多代理人 AI 系統之具體機制（不宣稱推進理論本身），提出 AI 對話三模式與便利貼 Build-on 機制，並提供正向科技接受度之初步證據。;As artificial intelligence (AI) increasingly automates procedural cognitive work， education is shifting toward design thinking， a human-centered， inherently small-group practice in which novices differ markedly from experts. Whether multi-agent AI can be an effective medium for it is therefore worth exploring.
This study developed and preliminarily evaluated the Design Thinking Studio (codename MindCrew). It uses the first diamond of the Double Diamond model (Discover/Empathise and Define) as its backbone， grounds the AI′s guidance in cognitive apprenticeship′s modeling， coaching， and articulation methods， employs three AI turn-taking policies (Cued， Round-Robin， and Open-Floor) as its collaboration core， and accumulates shared thinking context via a sticky-note Build-on mechanism. Forty-six learners each chose a policy and completed both stages under a self-service， asynchronous， between-subjects self-selection design; data came from a Technology Acceptance Model (TAM) questionnaire and open-ended feedback.
Technology acceptance was positive: perceived usefulness， ease of use， and intention to use all exceeded the scale midpoint significantly， as did the perceived usefulness of the Build-on mechanism and of both stages. Round-Robin was rated most favorably， but because policies were self-selected rather than randomly assigned， this difference remains preliminary and no causal claim is made. Open-ended feedback agreed with the quantitative results.
Its contributions are a theory-informed design translating three cognitive apprenticeship methods into multi-agent AI mechanisms (without claiming to advance the theory)， the three turn-taking policies with the Build-on mechanism， and preliminary evidence for future controlled work.
