
IEEE TOL – AI Tiered Scaffolding
It’s Official — and It Means More Than a Publication =)
I am delighted to share that this article, AI-Assisted Tiered Scaffolding for Programming Skill Development in a Problem-Based Game Development Environment, has been accepted and is now available as an Early Access article in IEEE Transactions on Education, one of the most respected journals in engineering education research.
This article is a reflection on a journey that began more than six years ago, shaped by curiosity about how students learn to program, deepened through theory and careful design, and now, with the arrival of generative AI, given a new and urgent dimension.
2020: Where It All Began (The UMP STEM Cube – LilEx)
The journey starts with a question we were asking in engineering education: how do we move students from passive learners to active makers? In 2020, we introduced the UMP STEM Cube (LilEx) a modular, hands-on learning tool strategically designed to elevate digital making skills in engineering programmes.

Grounded in Kolb’s Experiential Learning Theory, the STEM Cube was designed to close the gap between theory and practice. Students worked through programming, physical computing, and 3D modelling activities in a cycle of concrete experience, reflection, abstract conceptualisation, and active experimentation.
The findings were encouraging. Students who engaged with the STEM Cube demonstrated significant learning gains in programming competencies and physical computing skills. More than the numbers, their positive perceptions and growing confidence in digital making were what stood out, a signal that something in the design was working.
This work was published in IEEE Transactions on Education (2024): Assessing the Effectiveness of UMP STEM Cube as a Tool for Developing Digital Making Skill Sets.
2021: Designing the Scaffolding (Tiered and Intentional)
If 2020 was about the learning environment, 2021 was about the instructional design inside it. We turned to the question of how to structure activities so that learners are guided progressively, with support that gradually fades as competence grows.
The result was a tiered scaffolding framework, a carefully crafted sequence moving from full scaffolding toward independent work:
-
-
-
- Workout Programming step-by-step guidance through worked examples
- Debugging Specific Malfunctions identifying and fixing targeted errors in given code
- Semi-Completed Programming Tasks completing partially provided code
- New Programming Tasks applying concepts independently with minimal support
-
-

This design was tested in two contexts: graphical (block-based) programming and Python-based game development. Both studies confirmed that the tiered structure supported learners at different ability levels without overwhelming beginners or under-stimulating those who were ready to move ahead.
Related publications: Graphical programming study · Python game development study
2023–2026: What Happens When We Add AI .. Carefully =)
By 2023, generative AI tools were everywhere in education. Students were using them, often without guidance, with outcomes that raised real concerns: shallow understanding, passive reliance, and over-dependence on code generation without comprehension.
We did not want to ban AI. We wanted to embed it purposefully.
The central design question became: what does AI-assisted scaffolding look like when it is not left to chance? Our answer was to integrate AI interaction within the tiered scaffolding structure — not as a shortcut, but as a scaffold in itself. The AI interaction was framed around three modes:
-
-
- Code with AI, collaborating with AI to build and refine
- Code for AI, preparing clear prompts and specifications
- Code by AI, critically evaluating and adapting AI-generated output
-
The learning context was a Slider Game built in Pygame: a red enemy block falls from a random x-position; a green player block moves left and right to catch it. Eight structured activities (Acts) took students from orientation and game setup through movement logic, scoring, loop refinement, debugging, customisation, and reflection.

The study ran as a quasi-experimental design comparing an AI-assisted group (n = 60) against a control group (n = 55) following the same tiered scaffolding without AI access.
The Framework: Theory Doing Real Work
What holds this study together is not just the activities, but the theoretical architecture beneath them. Three foundations are woven together:
-
- Cognitive Apprenticeship provides the instructional sequence: Modeling → Coaching → Articulation → Reflection → Fading. Students are not simply given tasks; they observe expert thinking, receive guided feedback, articulate their own reasoning, reflect on their process, and progressively take ownership as scaffolding is withdrawn.
- AI Scaffolding is embedded within this sequence — not as an add-on, but as the mechanism through which coaching and feedback are delivered. AI suggestions are treated as materials to evaluate, not answers to copy.
- Problem-Based Learning anchors everything in an authentic, motivating context: building a working game. The game is not a gimmick. It provides the problem space within which every programming concept — from loops and conditionals to collision detection and scoring — earns its meaning.

The Learner–AI Programming Process cycle — Prompt → Code → Test → Debug → Reflect → (Prompt again) — describes how learners move through each activity, with AI as a responsive partner in that loop, never as a replacement for thinking.
What We Found
The numbers are hard to ignore. Students in the AI-assisted group achieved a mean learning gain of 25.51 points (SD = 13.68), compared to 6.90 points (SD = 6.16) in the control group. The effect size was large — Cohen’s d ≈ 1.86 — and the difference was highly significant: t(113) = 11.64, p < .001.
But the numbers only tell part of the story. Artifact and interaction log analysis revealed something more qualitative and perhaps more meaningful:
-
-
- Students in the AI-assisted group showed more systematic debugging — not random trial-and-error, but deliberate hypothesis-and-test cycles.
- Over time, there was a gradual shift toward critical evaluation of AI-generated suggestions, rather than passive acceptance. Students began questioning, adapting, and sometimes rejecting what AI produced.
- Self-efficacy and engagement were higher in the AI-assisted group — students expressed greater confidence and tackled problems more independently as the course progressed.
-
The core finding is that the educational value of AI depends not on the technology alone, but on the instructional framework in which it is embedded =)
On IEEE Transaction Peer Review: Why I Treasure the Process
I want to say something that is not always said out loud in academic circles: I am genuinely grateful for the peer review process, and I mean that without reservation =).
The reviewers’ comments on this manuscript were thorough, specific, and often challenging. There were moments of “oh”, where a reviewer pointed at a gap I had convinced myself did not exist, or pressed on a claim I had made too casually. The revision process was demanding. But it was also one of the most intellectually enriching parts of this entire journey.
Working through each reviewer comment forced me to revisit assumptions, sharpen my arguments, and in several instances, go back and strengthen the research itself , not just the manuscript. That is a rare gift. The final paper is better than the first submitted, and the credit for that belongs in large part to reviewers who clearly read with care and took the work seriously. This whole journey took almost 1 year =)
This is, I believe, why peer-reviewed publication carries the weight it does. It is not simply a stamp of approval. It is a process of scrutiny that when done well, makes the work more rigorous, more honest, and more useful to others who build upon it. I came away not just with an accepted paper, but with a deeper understanding of my own research. Thank you Editor and Reviewers.
What This Means and What Comes Next
We are at a moment in education where everyone is asking, what do we do with AI? Ban it, embrace it, ignore it and hope for the best?
This research offers a different kind of answer. The question is not whether to use AI, students are already using it. The question is how we design the instructional context in which that use happens. AI embedded within a thoughtful, theory-informed scaffolding framework can accelerate learning, deepen engagement, and develop more independent thinkers. AI used without structure can do the opposite.
The Slider Game, the tiered activities, the Cognitive Apprenticeship sequence, the Learner–AI Programming Process cycle, these are not prescriptions. They are a transferable design framework that educators can adapt for their own programming courses, their own students, their own contexts.
For me personally, this publication closes one chapter and opens another. Six years, three theoretical lenses, two learning environments, one framework, and now empirical evidence that the design works. I am proud of this body of work, and I am already thinking about where it goes next.
Thank you to my UMPSA STEM Lab colleagues, my students who gave their time and engagement to this research, the reviewers who sharpened it, and IEEE Transactions on Education for giving it a home =).
Read the paper: AI-Assisted Tiered Scaffolding for Programming Skill Development in a Problem-Based Game Development Environment IEEE Transactions on Education, Early Access, 2026.










































