How agentic AI brings interactivity to STEM learning

Electric and magnetic fields, wave propagation and signal transformations are foundational ideas in science and engineering, yet for learners encountering them for the first time they can be abstract. Traditionally, instructors have relied on static diagrams, mathematical derivations or simulations to convey these topics. Such tools, while helpful, are often inflexible and difficult to adapt to the specific needs of a class.

The emergence of agentic AI offers educators a new way to address this challenge. Unlike AI tools such as chatbots, which respond only to immediate instructions, agentic AI can plan, reason, generate code or simulations, and iteratively refine its outputs with minimal human intervention. So rather than simply delivering information, instructors can use AI agents to create bespoke simulations, visualisations and learning tools that make complex concepts more tangible. 

Students, meanwhile, gain the opportunity to develop computational thinking, modelling skills and problem-solving strategies in collaboration with intelligent systems. These tools also help them learn to work effectively with AI in technical domains.

Inspiration from dynamic STEM media channels

Students are increasingly drawn to YouTube channels such as Veritasium and 3Blue1Brown, which bring complex STEM topics to life. In these videos, variables shift in real time, graphs evolve and simulations respond seamlessly to the narration. Agentic AI offers educators a way to bring the advantages of interactivity into the classroom. Tailored visualisations allow students to manipulate parameters and instantly observe outcomes. Moreover, these AI-driven media can be refined based on student feedback and shared outside class, enabling students to engage with and revisit material at their own pace.

However, educators must also verify that the physics is represented correctly and that simulations behave as intended. 

From static diagrams to interactive exploration

In fields such as electromagnetism, students are often asked to reason about phenomena that cannot be directly observed. Concepts such as electric fields, magnetic flux or induced currents are typically introduced through equations and diagrams. While mathematically rigorous, this approach can make it difficult for students to develop intuition about how these systems behave.

• Balance human intuition with machine efficiency in scientific research
• Harnessing AI to expand scientific discovery
• Guard rails to mitigate ethical concerns for AI use

With agentic AI, instructors can rapidly prototype simulations that allow students to interact with these ideas. For example, an instructor teaching the Lorentz force might generate a simulation showing how a charged particle moves through combined electric and magnetic fields. By adjusting parameters such as field strength or particle velocity in real time, students can observe how trajectories change and begin to connect mathematical expressions with physical behaviour.

Rapid prototyping for teaching

One of the most powerful aspects of agentic AI in education is the ability to move quickly from idea to implementation. An instructor can outline the learning objective – for example, visualising electric field lines around multiple charges – and use an AI agent to generate the underlying code for a visualisation. They can then adjust parameters, add interactive controls or modify the visual representation to refine the output. This iterative process allows educators to tailor tools to teaching contexts rather than relying on generic simulations.

Such rapid prototyping also encourages experimentation in teaching. If a concept proves difficult for students to grasp, instructors can create alternative visualisations or demonstrations. Over time, this can lead to a richer set of teaching resources that are closely aligned with the needs of a particular course.

Importantly, these tools need not be complex to be effective. Even simple simulations that allow students to vary parameters and observe outcomes can significantly enhance conceptual understanding.

Ensuring accuracy and pedagogical integrity

In this partnership, the human user acts as a director while the AI functions as a competent worker. Yet even a capable AI requires clear instructions and feedback. And it can make mistakes. Code may contain errors, physics models may be simplified or misrepresented, and unexpected behaviours can arise. So careful validation and oversight when using AI-generated simulations in teaching remain important.

By explicitly addressing accuracy and pedagogy, instructors can harness the benefits of AI-generated simulations, interactivity, engagement and visual clarity without compromising the rigour essential to STEM education.

Best practice includes:

• validating the physics to ensure that equations, boundary conditions and interactions align with the intended concepts
• debugging and testing code, as well as reviewing and, if necessary, modifying AI-generated code to ensure reliable performance across parameter ranges
• ensuring that simulations reinforce learning objectives
• iterative refinement based on feedback to incorporate insights from students and colleagues to improve accuracy, clarity and usability.

Teaching students to work with AI

Beyond supporting instructors, agentic AI also provides an opportunity for them to rethink how students engage with computational tools.

Programming has long been an important skill in science and engineering education, but many students struggle with syntax and debugging. AI-assisted coding tools can help lower these barriers by generating working code that students can analyse, modify and extend. Rather than replacing the learning process, this approach shifts the emphasis towards understanding how models are constructed and how computational tools can be used to explore scientific questions. For example, students might be asked to use an AI agent to generate a simulation of wave interference or signal processing. Their task would then be to examine the code, verify that the underlying model is correct and experiment with different parameters. 

Through this process, students learn not only how to use computational tools but also how to critically evaluate AI-generated outputs.

Developing computational and scientific thinking

When used thoughtfully, agentic AI can transform coding from a purely technical exercise into a tool for scientific exploration. Students who previously spent most of their time troubleshooting syntax can instead focus on higher-level questions: what assumptions are built into the model? How do different parameters affect the system? Do the simulation results match theoretical predictions?

In this way, AI-assisted tools can reinforce core scientific skills such as modelling, hypothesis testing and verification. Students learn that while AI can generate code quickly, the responsibility for interpreting and validating results still lies with the human user.

The evolving role of the instructor

Instructors remain essential in designing meaningful learning experiences, guiding students in the interpretation of results and ensuring that technology is used critically and responsibly. Agentic AI can expand what is possible in the classroom. 

The challenge for educators is not simply whether to use AI in teaching, but how to use it in ways that promote understanding, curiosity and critical thinking. When applied thoughtfully, agentic AI has the potential to turn classrooms into environments where students actively explore complex systems rather than passively absorb information. In that sense, the most valuable contribution of AI to education may not be automation, but the new opportunities it creates for exploration and discovery.

Gerard Joseph Lim is a lecturer in physics and applied physics in the School of Physics and Mathematical Sciences at Nanyang Technological University, Singapore. 

This is an edited version of the post “Using agentic AI to build richer learning experiences in STEM”, first published on NTU’s Institute for Pedagogical Innovation, Research and Excellence blog.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.