Teaching Students to Use AI Wisely: New Research Highlights Opportunities and Risks in CS Education
Artificial intelligence (AI) is rapidly reshaping how software is built, and increasingly, how it’s taught. At the 2026 SIGCSE Technical Symposium, researchers from Georgia Tech’s School of Computing Instruction (SCI) will present how AI tools can be integrated into computing classrooms to support learning.
Evaluating AI Across Software Engineering Workflow
In their paper Benchmarking AI Tools for Software Engineering Education: Insights into Design, Implementation, and Testing, SCI faculty member Nimisha Roy, computer science (CS) major Oleksandr Horielko, and associate dean Olufisayo Omojokun explore how AI tools are reshaping software engineering workflows and what educators need to know about them.
The study builds on their previous course redesign work and benchmarks popular AI tools, including GitHub Copilot, ChatGPT, Claude, and Gemini, across key software engineering tasks: design, implementation, debugging, and testing. Five evaluators experienced in using AI tools for software engineering conducted controlled experiments with standardized prompts, measuring task speed, output accuracy, the amount of human correction required, and cross-file consistency.
The results show that AI tools can speed up boilerplate code generation, sketching UML diagrams, and other tasks, but they struggle to maintain test coverage and consistency across files, and to handle complex prompts.
“AI tools are often discussed as being either broadly beneficial or broadly harmful. Our results suggest a more nuanced picture,” Roy said. “The effectiveness of an AI tool depends heavily on where it is introduced in the workflow and on the level of scaffolding provided to students.”
Teaching How to Use AI
The study also highlights instructional challenges. Novice students may see AI as a shortcut to competence, while advanced students may automate tasks they’ve already mastered. Roy notes that students can over-trust AI outputs, assuming they are correct if they “look right,” which can undermine verification and debugging habits.
“Without explicit instruction in verification, refinement, and error detection, AI can unintentionally weaken students’ debugging and reasoning habits,” she said.
Omojokun emphasizes the rapidly evolving nature of AI tools.
“Something that's considered state-of-the-art at the beginning of a semester can easily become old news before the end of the term. Instructors should continually survey the technological landscape to ensure that the concepts they are teaching aren't contradictory to what's current,” he said.
The researchers offer practical guidance for instructors looking to integrate AI effectively into software engineering courses, helping students benefit from AI while developing strong coding and problem-solving skills.
How AI Is Shaping Student Learning in Introductory CS
Extending this work on AI in computing education, two SCI posters also examine how AI tools influence student learning in undergraduate CS classrooms.
The first poster, To Tell or to Ask? Comparing the Effects of Targeted vs. Socratic AI Hints, by graduates and research assistants Zhixian Liding, Michael Osmolovskiy, and Harshith Lanka, along with SCI faculty Ronnie Howard, Nimisha Roy, and Rodrigo Borela, examine how AI-generated hint styles affect students.
In a randomized trial involving 178 CS students, those who received Socratic, question-based hints spent more time and made more attempts to solve problems. However, they showed no clear long-term learning gains during the study period. The findings highlight the trade-offs between encouraging productive struggle and maintaining efficiency.
The second poster, AI-Augmented Instruction: Real-Time Misconception Detection, is by Liding, Osmolovskiy, Lanka, Roy, and Borela.
It introduces a system that uses AI to classify and cluster student coding errors in real time. An instructor dashboard surfaces common misconceptions across a class, enabling more targeted, timely instructional support in large courses.
More to Explore
In addition to these AI-focused efforts, SCI researchers are presenting other CS education work at SIGCSE 2026, which runs from February 18 to 21 in St. Louis, Missouri. Explore the full lineup here.
