The Problem: Why Reflection Needed Rethinking 

Perspective: Timothy Davies 

Renowned educational philosopher and psychologist, John Dewey, is famously quoted in saying “We do not learn from experience, we learn from reflecting on experience”. In experiential learning contexts (e.g. clinical placements), assessing student’s ability to reflect has long been a challenge. Students often struggle to engage deeply with reflection, especially when tasks focus more on polished writing than on the thinking process itself. We also observed gaps in collaborative project skills with limitations in how reflection was scaffolded and assessed within units of study themselves. Moreover, in the age of GenAI, students are able to produce written reflections with exceptional accuracy with potential for agents to hallucinate and generate interactions/observations that don’t exist. These issues aligned with broader assessment reform with Liu and Bridgeman’s two-lane approach being up taken by our University. This model acknowledges the power of AI to assist student learning in open environments, with assurance of learning occurring in secure environments. 

An idea emerged: could AI help scaffold student learning during placements by removing the element of creating a written narrative and focusing on the process of deep reflection? 

Conceptualisation and Prototyping: Piloting the Reflective Agent 

Perspective: Timothy Davies 

The initial prototype, built in Cogniti.ai, was structured around reflective models that may be used in various disciplines across a higher education institution. It aimed to guide students through structured prompts aligned with these frameworks with follow-up questions specific to student inputs. Early trials revealed that students engaged with the agent in two themes: curiosity/desire to learn and negative feelings such as frustration and disappointment. Educators who used the agent were very positive, with one in Physiotherapy stating that “Students produced deeply insightful reflections after using the agent to guide their thinking. It helped them learn the reflection process, which they then applied independently to complex clinical scenarios involving cultural diversity.” However, underlying technical and pedagogical constraints impacted its ability to be reliably implemented at scale. Specifically, students found the model selection unclear, and their responses suggested a need for more nuanced support. 

We realised that deeper design support was needed in how the agent was prompted and how it engaged students in authentic, process-oriented reflection. 

Designing with Purpose: AI as Reflective Coach in Placement Learning 

Perspective: Ruby Nguyen  

When Tim and I began discussing his reflective agent in Cogniti.ai, we found shared territory. My previous work with pre-service teachers and school placements had wrestled with similar challenges: guiding meaningful reflection in placement contexts where students are under-supported, emotionally taxed, and juggling shifting expectations. Our question was: How do we design for reflection as evolving practice while making it assessable, scaffolded, and ethically AI-supported? 

Tim's original agent walked students through reflective models via structured prompts, but student responses revealed needs for more pedagogical precision. I revised the system message, reorienting the agent as a Socratic reflective coach. Rather than generating a reflection or leading toward "model answers," it asks well-sequenced, open-ended questions, responds to thinking depth, and prompts further articulation. 

The agent draws from established models (Rolfe, Driscoll, Gibbs, Kolb, Bain's 5R) through clear structure: students choose from frameworks or default to "What? → So what? → Now what?". We wanted diverse students to relate to the agent in an informed and autonomous way, without placing the burden of theoretical mastery on them during an already demanding learning activity. 

Reflection ends with what students do with those insights. Another crucial conversation we had focused on the action planning stage. Tim raised a thoughtful critique of the SMART goals framework around its tendency to reduce student thinking to formulaic or performance-oriented responses. We, therefore, reworked goal-setting using principles from self-determination theory, encouraging students to consider not only what they wanted to do, but why, how, and on what terms.  

Just as important was the student use of the agent in alignment with assessment. Tim's rubric rewarded "depth" while implicitly favouring polished writing over process. Our collaboration on the revised agent surfaces processual qualities such as observing, analysing, linking, planning, recognising progression over perfection. This led Tim to reconsider rubric design itself.  

What we have built is a structure for thinking; a reflective encounter shaped by two educators' conversations and pedagogical commitments. The real work lies in how students engage with it, and how we continue adjusting our practices in response. 

Reflections on the Co-Designed Collaborative Process 

Perspective: Timothy Davies 

Working with central educational innovation teams challenged me to think differently about AI as a pedagogical partner. As the project evolved, my own thinking shifted. I began to see the difference between tweaking prompts and redesigning learning with a perspective that differed from my discipline. The strength in this approach allowed my disciplinary expertise, combined with educational insight to be independently interrogated in a collaborative approach to move forward with this problem within a structured ethical framework. 

Maintaining student-centredness in a highly structured tool required constant reflection and collaboration. Our collaboration helped surface the pedagogical values embedded in design decisions, how we scaffold autonomy, assess depth, and support emotional and cognitive labour in placement learning. 

Blog Authors:

• Dr Tim Davies - University of Sydney

• Dr Ruby-Ngoc Nhu Nguyen - University of Sydney

Banner Image source: chenspec, Pixabay