This paper presents an evidence-based instructional methodology for mathematics education that combines traditional instruction with AI-assisted problem solving using ChatGPT integrated with Wolfram Alpha. The approach was implemented in two first-year engineering topics, integral calculus and graphical analysis, to evaluate its effectiveness across different mathematical contexts. The instructional design included a traditional session and an AI-assisted session. Students' perceptions were measured through a Likert-scale survey applied to 49 participants, assessing perceived difficulty, attention, confidence, and ease of error correction. Statistical analysis showed higher perception scores for the AI-assisted activity, with significant improvement in the first implementation ($p=0.024,95 {\%}$ confidence) and a positive trend in the second ($p=0.053,90 {\%}$ confidence). Additionally, 90% of students rated the AI feedback as helpful, and 84% reported increased confidence. Future work will focus on analyzing whether this perception translates into measurable learning gains. These results suggest that AI-assisted activities can effectively complement traditional instruction.