Case Study: MagicSchool Custom AI Tool Development
MagicSchool Ambassador Program / Broward County Public Schools, Fall 2026

The Program

The MagicSchool Ambassador program has included open nomination cycles in the past. The Fall 2026 cohort was different. Participation required a direct nomination from a MagicSchool employee, making it a more selective group. I was nominated and selected to participate in a structured two-week program that included completing course modules, building custom AI tools within the MagicSchool platform, and creating and sharing resources with the broader ambassador community.

The Approach

Rather than building tools for the sake of building them, I started with real pain points. Every tool I designed came from something educators, administrators, or my Canvas admin team had mentioned as a recurring frustration or unmet need. My role in each case was to write the underlying instructions, define the input fields and output structure, and build in quality rules to keep the tool focused and reliable.

The Canvas Global Announcement Generator was the one exception. That one came directly from my work as a district Canvas LMS administrator, where drafting clear, consistent announcements for 200,000 plus users is an ongoing operational need.

Once the instructions were written, I tested each tool through multiple rounds of prompting and used MagicSchool’s built-in prompt assistant to review the final version, ensuring the instructions were thorough and optimized for best use. Testing also revealed edge cases that initial instructions did not anticipate. The Cold Reads Generator is a good example. When I tested it with a chemistry prompt, the output included symbols that would have been unreadable in a fluency assessment context. Students taking a fluency assessment need to read words aloud, not interpret chemical notation. I went back into the instructions, added explicit rules about avoiding symbols, and retested until the output was reliable across content areas. That kind of iterative review is built into how I approach any tool or resource I build, because the first version is rarely the final version. 

The Tools

• AI Literacy PD Planner: Built using MagicSchool’s Step by Step feature, this multi-step tool guides users through planning AI literacy professional development for teachers. It embeds digital citizenship, transparency, bias review, and privacy principles into the planning process, helping users think through stronger and more responsible PD options rather than defaulting to surface-level AI training.

• Cold Reads Generator: Addresses a recurring classroom need for fluency assessment material across content areas. Teachers input subject, grade level, standards, word count, and target vocabulary, and the tool produces a content-aligned fluency passage. It removes the manual work of sourcing or writing cold reads while keeping the output tightly tied to instructional context.

• Canvas Global Announcement Generator: Built to support the district Canvas admin team in drafting global announcements for staff. Announcements need to be clear, consistent, and appropriately toned for a large and diverse audience. This tool helps streamline that process.

• Staff Meeting Engagement Strategy Builder: Designed for administrators who want to move beyond passive staff meetings. The tool generates personalized engagement strategies to help administrators plan active, collaborative sessions based on their specific context and goals.

• Meeting Agenda and Minutes Synthesizer: Takes rough meeting notes and transforms them into a structured agenda, objective minutes, action items with owners and deadlines, a summary paragraph, and an optional follow-up email. A guardrail built into the system instructions prevents the tool from adding any content not present in the source notes, keeping the output accurate and trustworthy. Of course, it is important to always review AI-generated content.

What This Work Demonstrates

Building effective AI tools requires more than prompting. It requires understanding the user’s actual workflow, defining what good output looks like, anticipating where the tool might go wrong, and writing instructions precise enough to prevent it. Each of these tools reflects that design thinking process applied to a real need in an educational context.

The tools are shared within the MagicSchool Ambassador community and available for other MagicSchool Ambassadors to access and distribute. Additional opportunities to present this work to broader audiences may follow as the program continues.