The integration of artificial intelligence into educational settings has evolved beyond general-purpose tools to encompass sophisticated, educator-controlled AI agents designed specifically for learning enhancement. This development represents a significant shift towards pedagogically-focused AI implementation, where educators maintain control over AI interactions while leveraging the technology’s power to support student learning outcomes. The emergence of platforms like Cogniti and similar educational AI tools demonstrates how custom AI design can address specific teaching challenges whilst maintaining educational integrity and purpose.
The Foundations of Custom Educational AI Design
Understanding AI Teaching Assistants
Custom AI agents in education function as sophisticated digital teaching assistants that can be precisely tailored to meet specific pedagogical objectives. Unlike generic AI tools, these agents operate within defined parameters set by educators, ensuring alignment with learning outcomes and institutional standards. The key distinction lies in their ability to be “steered” through carefully crafted system messages that define their behaviour, scope, and interaction patterns.
The concept draws parallels to onboarding a human teaching assistant, where specific instructions guide the assistant’s role within the course. Just as educators would provide clear guidance to human TAs regarding their responsibilities, interaction styles, and educational objectives, custom AI agents require similar detailed briefings through system messages. This approach ensures that AI interactions remain educationally meaningful and aligned with broader pedagogical goals.
Research indicates that educators who utilise AI for strategic input rather than merely content generation achieve significantly better educational outcomes. This finding reinforces the importance of thoughtful AI design that considers the entire learning ecosystem rather than focusing solely on content production.
The System Message Framework
The effectiveness of custom educational AI agents largely depends on the quality of their system messages – the foundational instructions that govern AI behaviour. These messages serve as the primary mechanism through which educators maintain control over AI interactions, defining not just what the AI should do, but how it should approach educational tasks.
The system message represents the educator’s pedagogical intentions translated into actionable AI instructions. This translation process requires careful consideration of learning objectives, student needs, and desired interaction patterns. Effective system messages establish clear boundaries while maintaining sufficient flexibility to accommodate diverse student inquiries and learning styles.
The development of robust system messages involves understanding both the technical capabilities of AI systems and the nuanced requirements of educational contexts. Educators must balance specificity with adaptability, ensuring that AI agents can respond appropriately to unexpected student questions while maintaining focus on core learning objectives.
The RTRI Method for AI Agent Design
Role Definition and User Context
The Role-Task-Requirements-Instructions (RTRI) framework provides a structured approach to designing effective educational AI agents. The first component, Role, defines the AI agent’s identity within the educational context. This might encompass various roles such as writing tutor, problem-solving guide, discussion facilitator, mentor, coach, or subject-specific expert.
Equally important is defining the user role – typically students at specific levels or with particular learning needs. This dual role definition enables the AI to tailor its communication style, complexity level, and approach to match the intended audience. For instance, an AI agent designed for first-year students would adopt a different tone and provide more foundational explanations compared to one designed for advanced learners.
The role definition extends beyond simple functional descriptions to encompass pedagogical approaches. An AI agent designed as a Socratic tutor would engage differently than one functioning as a direct instructor or collaborative learning partner. This distinction becomes crucial in maintaining consistency with broader pedagogical strategies.
Task Specification and Requirements
The Task component of the RTRI framework requires explicit definition of what the AI agent is expected to accomplish. This might involve facilitating role-play scenarios, guiding students through complex problems, providing feedback on drafts, or acting as a simulator for real-world situations. Task specification benefits from granular detail, as specificity typically correlates with more effective AI performance.
Requirements outline the standards and criteria that govern AI performance. These might include depth of explanations, use of specific terminology, alignment with particular learning approaches, or adherence to institutional guidelines. Requirements serve as quality control mechanisms, ensuring that AI outputs meet educational standards and expectations.
The requirements component also addresses ethical considerations and boundaries. For educational AI agents, this might include prohibitions against providing direct answers when the learning objective involves student discovery, guidelines for handling sensitive topics, or protocols for referring students to human support when appropriate.
Instructions and Operational Guidelines
Instructions provide detailed operational guidelines for AI agent behaviour. These might specify response tone, use of examples, steps to follow in particular situations, or rules that must govern all interactions. Instructions translate pedagogical philosophy into practical AI behaviour, ensuring consistency across student interactions.
Effective instructions often include conditional logic – guidelines for different scenarios that might arise during student interactions. For example, instructions might specify how to respond when students ask off-topic questions, how to handle requests for direct answers, or how to adapt explanations based on apparent student understanding levels.
The instructions component also addresses integration with broader learning activities. This might include guidelines for referencing course materials, connecting discussions to learning objectives, or preparing students for upcoming assessments or activities.
Implementation Strategies and Best Practices
Pedagogical Alignment and Learning Objectives
Successful implementation of custom educational AI agents requires clear alignment between AI design and specific learning objectives. This alignment process involves three key steps: defining educational objectives, translating objectives into AI-understandable language, and balancing specificity with flexibility.
The process begins with articulating precise learning outcomes that the AI agent should support. These might include developing critical thinking skills, understanding specific concepts, or mastering particular problem-solving approaches. Clear objective definition provides the foundation for all subsequent AI design decisions.
Translating educational objectives into AI instructions requires breaking down abstract pedagogical concepts into concrete, actionable directives. For example, fostering critical thinking in historical analysis might translate into specific instructions for encouraging multiple perspectives, questioning source credibility, and drawing connections between historical and contemporary issues.
The balance between specificity and flexibility ensures that AI agents can adapt to individual student needs while maintaining focus on core objectives. This balance prevents overly rigid interactions while ensuring consistent support for learning goals.
Resource Integration and Content Management
One significant consideration in custom AI design involves the integration of educational resources. While educators might assume that uploading extensive materials enhances AI knowledge, research suggests that this approach can sometimes limit effectiveness. AI systems typically perform keyword searches within uploaded resources rather than comprehensively assimilating information, potentially leading to fragmented or out-of-context responses.
The most effective approach often involves leveraging the AI’s existing knowledge base while integrating essential, unit-specific information directly into system messages. This might include assessment rubrics, specific instructions, key concepts, or unique course requirements. Selective resource inclusion, focusing only on materials that are absolutely crucial and unique to specific educational contexts, tends to produce better outcomes.
This approach recognises that AI systems often perform more effectively when guided by clear instructions rather than overwhelmed with extensive documentation. The analogy of providing a human teaching assistant with essential guidance rather than exhaustive materials applies equally to AI agent design.
Iteration and Quality Assurance
Effective custom AI design requires ongoing refinement based on actual student interactions and feedback. This iterative approach begins with small-scale pilots, allowing educators to observe AI performance and identify areas for improvement before full implementation.
Monitoring student interactions provides valuable insights into AI effectiveness and areas requiring adjustment. This monitoring might involve reviewing conversation histories, analysing student feedback, and identifying patterns in AI responses that may not align with pedagogical intentions.
Regular quality assessment focuses on ensuring AI responses maintain accuracy, relevance, and alignment with educational objectives. This process might reveal recurring issues that require system message modification or highlight successful interaction patterns that could inform future designs.
The iterative refinement process acknowledges that effective AI design develops over time through continuous improvement based on real-world usage data and educational outcomes.
Diverse Applications in Educational Contexts
Subject-Specific Implementation
Custom AI agents demonstrate particular effectiveness when designed for specific subject areas and learning contexts. Examples include Socratic tutors for biology that guide students toward understanding through questioning rather than direct instruction, role-play facilitators for historical scenarios, and problem-solving guides for mathematical concepts.
Research in engineering education demonstrates successful AI integration through custom chatbots acting as industry consultants in safety case studies. These applications show high student engagement rates, with 83% of students utilising AI consultants and 71% engaging with AI interview tools. Such implementations suggest that students often prefer AI interactions for certain types of learning support, particularly when social anxiety might inhibit direct questioning.
The effectiveness of subject-specific AI agents stems from their ability to maintain focus on disciplinary knowledge and methods while providing consistent, patient support for student learning. This specialisation allows for deeper integration with course content and more sophisticated pedagogical approaches.
Multilingual and Accessibility Applications
AI agents demonstrate significant potential for supporting diverse student populations, including those studying in non-native language environments. Automated captioning and language support features can enhance accessibility while maintaining educational quality for all students.
The global nature of higher education increasingly requires tools that can support students from diverse linguistic and cultural backgrounds. Custom AI agents can be designed to provide culturally responsive support while maintaining academic standards and expectations.
These applications highlight the potential for AI to address educational equity concerns by providing consistent, high-quality support regardless of student background or individual learning needs.
Assessment and Feedback Integration
Custom AI agents show particular promise in assessment design and feedback provision. When integrated thoughtfully with backward design principles, AI can support the creation of formative assessment systems that truly serve learning objectives.
The integration of AI into assessment represents more than technological enhancement – it offers opportunities to reimagine evaluation approaches and provide more responsive, personalised feedback to students. This integration requires careful consideration of learning objectives and assessment purposes to ensure that AI support enhances rather than replaces meaningful educational evaluation.
Effective assessment integration focuses on using AI to support student learning processes rather than simply automating grading or content generation. This approach aligns with educational research emphasising the importance of formative assessment in supporting student achievement.
Challenges and Considerations
Technical and Pedagogical Limitations
Despite their potential, custom educational AI agents face several implementation challenges. Technical limitations include occasional inaccuracy in responses, particularly for numerical problem-solving tasks. Research in mechanical engineering education reveals that while AI tools perform well with theoretical questions, they struggle with complex calculations and deep conceptual understanding.
Pedagogical challenges include the need for substantial training for both educators and students to effectively utilise AI tools. The requirement for robust technical infrastructure and ongoing support can present barriers for some institutions, particularly those with limited resources.
These limitations suggest that successful AI implementation requires careful planning, adequate training, and realistic expectations about AI capabilities and constraints.
Ethical and Educational Integrity Concerns
The integration of AI into educational settings raises important questions about academic integrity and student learning. Concerns about student dependency on AI tools and potential impacts on problem-solving skill development require careful consideration in AI design and implementation.
Educational institutions must develop clear guidelines for AI use that support learning objectives while maintaining academic standards. These guidelines should address appropriate use cases, limitations, and ethical considerations specific to educational contexts.
The development of AI literacy among students becomes crucial for ensuring that AI tools enhance rather than replace critical thinking and learning skills. This requires intentional instruction about AI capabilities, limitations, and appropriate use in academic contexts.
Implementation and Support Requirements
Successful custom AI implementation requires significant institutional commitment to training, support, and ongoing development. Survey research from Australian higher education institutions reveals varying levels of preparedness and understanding regarding AI implementation in educational contexts.
The need for specialised knowledge in both educational technology and pedagogical design can present challenges for institutions seeking to implement custom AI solutions. This requirement may necessitate additional professional development or specialised staff roles.
Long-term sustainability requires ongoing attention to AI agent performance, student feedback, and evolving educational needs. This iterative improvement process demands dedicated resources and institutional commitment to continuous enhancement.
Conclusion
The design of custom AI agents for educational purposes represents a significant advancement in educational technology, offering unprecedented opportunities for personalised, responsive learning support. The success of platforms like Cogniti and similar tools demonstrates that when educators maintain control over AI design and implementation, these technologies can meaningfully enhance student learning outcomes whilst preserving educational integrity.
The RTRI framework provides a practical approach to AI agent design that balances technical capabilities with pedagogical requirements. Through careful attention to role definition, task specification, requirements articulation, and detailed instructions, educators can create AI agents that truly serve learning objectives rather than simply automating existing processes.
The evidence suggests that the most effective educational AI implementations focus on strategic integration that enhances rather than replaces human teaching expertise. When designed with clear pedagogical intentions and implemented through iterative refinement processes, custom AI agents can provide valuable support for diverse student populations whilst maintaining the human connections essential to meaningful education.
Future development in this area should continue to emphasise educator control, pedagogical alignment, and student-centred design principles. As AI technology continues to evolve, the focus must remain on how these tools can best serve educational objectives and support student success in increasingly complex and diverse learning environments.
References
Cogniti. (2023). AI agents designed by teachers. https://cogniti.ai
Cogniti. (2024, August 23). How does Cogniti work? Cogniti Documentation. https://cogniti.ai/docs/how-does-cogniti-work/
Educational Innovation Sydney. (2023, June 6). Prompt engineering for educators – making generative AI work for you. Teaching@Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/prompt-engineering-for-educators-making-generative-ai-work-for-you/
Educational Innovation Sydney. (2023, December 8). How Sydney educators are building ‘AI doubles’ of themselves to help their students. Teaching@Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/how-sydney-educators-are-building-ai-doubles-of-themselves-to-help-their-students/
Educational Innovation Sydney. (n.d.). Letting educators take control of generative AI to improve learning, teaching and assessment. Teaching@Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/letting-educators-take-control-of-generative-ai-to-improve-learning-teaching-and-assessment/
Educational Innovation Sydney. (2024). How to design custom AIs that can propel learning. Teaching@Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/how-to-design-custom-ais-that-can-propel-learning/
Education Services Australia. (2023, August). AI in Australian education snapshot – Principles, policy and practice. https://www.esa.edu.au/docs/default-source/default-document-library/ai-in-australian-education-snapshot—principles-policy-and-practice_august-2023.pdf
ESKilled. (2025, June 26). Comprehensive guide to implement AI in RTO curriculum design. AI Course Creator. https://aicoursecreator.eskilled.io/blog/guide-to-implementing-ai-in-rto-curriculum-design/
Hong, S., Zheng, X., Chen, J., Cheng, Y., Wang, J., Zhang, C., Wang, Z., Yau, S. K. S., Lin, Z., Zhou, L., Ran, C., Xiao, L., Wu, C., & Schmidhuber, J. (2024). Evolution of AI in education: Agentic workflows. arXiv preprint arXiv:2504.20082. https://arxiv.org/pdf/2504.20082.pdf
Jiang, Y.-H., Li, R., Zhou, Y., Qi, C., Hu, H., Wei, Y., Jiang, B., & Wu, Y. (2025). AI agent for education: von Neumann multi-agent system framework. arXiv preprint arXiv:2501.00083. https://arxiv.org/html/2501.00083v1
Islam, S., & Breuer, F. (2025, March 13). Harnessing AI for smarter teaching and personalised learning. University of Newcastle. https://www.newcastle.edu.au/newsroom/college-of-engineering-science-and-environment/harnessing-ai-for-smarter-teaching-and-personalised-learning
Massachusetts Institute of Technology. (2024, March 1). Where to start: Backward design. Teaching + Learning Lab. https://tll.mit.edu/teaching-resources/course-design/backward-design/
MeraTutor. (2025, June 11). Beyond answers: AI chatbots and the rise of Socratic learning. MeraTutor AI Blog. https://www.meratutor.ai/blog/socratic-learning/
Ortega-Ochoa, C., Kumar, A., & Wu, Y. (2024). Exploring how teaching assistants powered by AI can understand and support students. British Educational Research Association Blog. https://www.bera.ac.uk/blog/exploring-how-teaching-assistants-powered-by-ai-can-understand-and-support-students
Parolin, T. (2025, July 15). AI and its impact on education. Australian Public Service Academy. https://www.apsacademy.gov.au/news/ai-and-its-impact-education
Pedagogie Digitala. (n.d.). AI agents in education: An early systematic review of emerging roles, potential, and limitations. https://pedagogie-digitala.ro/ai_agents_in_education_early_systematic_review/
Singh, A., Kamra, N., Patel, M., Patel, D., & Chen, M. (2024). Multi-agent system for automated essay scoring: A proof-of-concept application. In Proceedings of the International Conference on Machine Learning Applications (pp. 1-8). IEEE.
Tarabishy, A. (2024, September 10). The Socratic method in the age of artificial intelligence. International Council for Small Business. https://icsb.org/ayman-tarabishy/the-socratic-method-in-the-age-of-artificial-intelligence/
Turnitin. (2023, February 13). Backward design: Empowering students through assessment. Turnitin Blog. https://www.turnitin.com/blog/backward-design-how-can-meaningful-assessment-empower-students
University of Queensland. (n.d.). Academic AI – 11 Socratic tutoring. Psychology Department. https://www.psy.uq.edu.au/~uqjtange/academic_ai/t_socratic_tutoring.html
Vorobyeva, D., Tonbulo?lu, ?., & Abbas, A. (2025). Personalized learning through AI: Pedagogical approaches and critical insights. Contemporary Educational Technology, 17(2), ep574. https://www.cedtech.net/download/personalized-learning-through-ai-pedagogical-approaches-and-critical-insights-16108.pdf
WGU Labs. (2023, May 1). Move curriculum forward with backward design. https://www.wgulabs.org/posts/moving-curriculum-forward-with-backward-design
Wu, X., Xie, H., Zou, D., & Wang, F. L. (2024). Multi-agent systems for automated essay assessment: Design principles and implementation strategies. Computers & Education, 198, 104-118.
AI Author Statement
This experimantal article was researched and written using Perplexity AI and the Labs function. The author acknowledges the use of AI assistance in gathering information, synthesising research findings, and structuring the content presented in this review.
Leave a Reply