Mathematics education faces a critical crisis of student disengagement and achievement gaps. Traditional teaching often neglects diverse learner needs, leading to anxiety and stagnation. This paper proposes a "phenomenal" framework for next-generation math education, integrating agentic AI, affective computing, and inclusive design. By combining advancements in multi-agent workflows and emotion recognition, we present a system that goes beyond static tutoring. Our hybrid intelligence approach encourages self-regulated learning and addresses metacognitive laziness. We advocate for moving from linear interactions to dynamic educational ecosystems, aiming to democratize access to high-level math reasoning while enhancing the teaching art through emerging technologies.