Artificial intelligence (AI) is rapidly reshaping architectural practice, yet its pedagogical integration remains uneven, especially in contexts where digital infrastructure, faculty training, and curriculum reform are still developing. Focusing on Bangladesh, this paper examines how AI-powered design tools can be incorporated into undergraduate architecture education in ways that expand learning while protecting foundational design competencies. The study adopts a structured literature review, analyzing scholarship on AI in architectural practice, architecture and design education, and technology-enhanced learning. Sources were selected for their relevance to four themes: types of AI tools used in design, pedagogical opportunities, educational risks, and models for curriculum integration. The review shows that tools such as Midjourney, DALL-E, Grasshopper, Dynamo, and Autodesk Forma can support broader creative exploration, faster iteration, early exposure to contemporary professional workflows, and engagement with computational and performance-driven design. At the same time, the Bangladeshi context presents significant challenges, including over-reliance on machine-generated outputs, possible weakening of freehand and spatial reasoning skills, unresolved authorship and academic integrity questions, uneven faculty preparedness, and inequitable access to hardware, internet, and paid platforms. In response, the paper proposes a four-phase pedagogical framework: foundational manual skill development in Year 1; supervised AI-assisted ideation and image critique in Year 2; performance-based generative and parametric applications in Year 3; and ethically reflective, professionally oriented AI integration in senior studios. By framing AI adoption as a pedagogical and institutional issue rather than a purely technical one, the paper offers a context-sensitive model for balancing innovation, equity, and disciplinary values in architecture education.