Colleague.AI: An Educational Platform with AI Assistants as Knowledgeable Colleagues for Educators and Friendly Buddies for Learners

• In Event: Innovating K–12 Teaching and Learning with Generative AI: Insights From IES-funded National R&D Centers

Abstract

This poster presents the design rationale, core capabilities, and early evaluation framework for Colleague AI, an AI-powered platform co-designed with educators to support core instructional tasks in K-12 education. The objective is twofold: (1) to reduce teacher workload through automation of lesson planning, assessment, and content customization, and (2) to enhance student learning opportunities through AI-mediated tutoring and engagement. In developing and researching the use of generative AI (genAI) tools in education systems, we focus particularly on their implications for instructional quality, teacher trust in AI-generated content, and the conditions required for scalable, responsible implementation.
The work is grounded in the principle that AI tools should augment, not replace, educator judgment. We conceptualize Colleague AI as a third agent in the classroom, working with teachers and students. The platform is based on co-design research (Sarkar et al., 2025a), as we work actively with educators and school districts to design, test, and iterate on core platform features. The educational AI assistant integrates Open Educational Resources (OER) and aligns with curriculum standards to deliver high-quality, adaptable content.

In this poster we will present key features of the platform including the interface for teachers to interact with the genAI assistant (Figure 3) and the classroom page where teachers can create assignments, and design genAI-based chatbot conversations for their students to engage with on specific topics (Figure 4). We will also present findings from research studies about how teachers are engaging with and evaluating the platform.

Data includes teacher co-design interviews, usage surveys, and platform administrative log data from pilot users users. All research has been conducted with IRB review and approval. The lesson plan generation tool is grounded in pedagogical best practices, and an A-B comparison study showed that teachers rated AI-generated lesson plans as comparable in quality to those created by human content experts (Sarkar et al 2025b). AI-evaluation of student feedback is based on rubrics that teachers control on assessments that teachers assign (Tian et al 2025a). In a co-design study of classroom features, teachers assessed the quality of generated feedback on student work, with 58% of teachers reporting that the feedback was useful for teachers or students (see Figure 5, Tian et al 2025b).

This work contributes to emerging scholarship on applied genAI integration in education by offering a concrete, scalable example of how genAI can be aligned with instructional values and institutional constraints. By foregrounding educator agency in the platform's design and evaluation, Colleague AI offers a counterpoint to techno-solutionist narratives, suggesting more grounded pathways for genAI adoption in public K-12 systems. Future work includes quasi-experimental evaluation of learning outcomes in a year-long pilot evaluation across varied educational contexts.

Authors

• Alex Liu, University of Washington

• Lief Esbenshade, University of Washington

• Shawon Sarkar, University of Washington

• Zewei Tian, University of Washington

• Zachary Zhang, Colleague.ai

• Kevin He, Colleague.ai

• Min Sun, University of Washington