While educational repositories offer specialised metadata, teachers prefer general-purpose web search engines due to better results and familiar interfaces. But web search engines do not surface the educationally relevant information required for effective resource evaluation. This creates a key problem: integrating educational information into web search without disrupting existing workflows. Through a design study, we investigated how AI-enhanced interfaces can support teacher-AI collaboration in resource selection. After observing current search practices (N = 9), we identified opportunities for AI assistance and developed a browser extension to augment web search interfaces with adaptive educational summaries. We designed the interface through an iterative refinement stage with four participants, followed by a final evaluation using think-aloud protocols, search logs, questionnaires and semi-structured interviews (N = 8). Our evaluation revealed some guidelines for teacher-AI educational partnerships in search: a.) interfaces should support a two-stage decision process combining quick metadata-based decisions followed by detailed pedagogical assessment b.) pedagogical details must be presented selectively to balance insight with cognitive load, and c.) interfaces must support customisation as teachers have distinct resource evaluation approaches. These findings could inform the design of AI assistance systems that complement teacher expertise and fit into their existing workflows.