Beyond Automation: How Agentic AI Can Revolutionise Bangladesh’s Science Classrooms

Step into a science classroom almost anywhere in Bangladesh, from a packed lecture hall in Dhaka to a school in a remote sub-district, and you will see the same scenarios: too many students, too little lab equipment and an education system that rewards memorising over wondering. Curiosity does not survive for long.

In Dhaka and other big cities, students are squeezed into crowded rooms and spend their evenings at coaching centres, where the aim is to master exam patterns, not to understand science. In remote sub-districts, the gaps are more basic. There are not enough trained science teachers, the lab cupboards hold broken or missing equipment, and the electricity supply and internet connection are not stable. Earlier digital fixes like YouTube tutorials, online quizzes and simple chatbots helped a little because they mostly reached students who already had decent devices and steady connections, and they rarely got anyone to think like a scientist.

Agentic AI could change that, as long as we bring it in with care. While most AI tools just wait for a prompt and answer it, agentic AI works towards a goal: it plans, reasons and carries out several steps on its own. In a STEM setting, that means it does not just solve a physics problem or summarise a biology chapter. It can act like a lab assistant, a research partner and a mentor all at once. It can design an experiment, test a hypothesis, flag strange results in your data, and push back when your reasoning does not hold up.

How it fits into the system matters as much as what it can do. In urban schools and universities, agents can be built into existing labs, coursework and research projects, so students move from receiving information to running their own inquiry. In rural areas, the same ability has to arrive in a lighter form: Bangla-language agents that run on low-cost smartphones or shared school computers, work offline or on a weak signal, and sync whenever a connection appears. Community spaces like Union Digital Centres could reach students whose schools have no devices at all. And throughout, the agent should support teachers instead of working around them. It can help one science teacher handle a huge class, prepare simulations and give individual feedback, but only if teachers are trained and involved from the start. Done well, a student in a remote upazila and a student in Dhaka could practise the same kind of scientific thinking, even if their equipment looks very different.

For a country trying to modernise its education system, that is a real chance to skip over some infrastructure gaps. Some universities are already moving in this direction. Universities like the American International University-Bangladesh (AIUB), with its strength in computer science and engineering, is a natural place to pilot these tools in science and engineering courses. If it does, students could shift from simply receiving technical information to actually running scientific inquiries themselves.

Picture a chemistry student whose lab has no reagents. An AI agent can build a detailed virtual simulation, change the conditions as the student watches, and ask them to explain a reaction they did not expect. In computer science or physics, several agents could work together like a small research team, with one proposing hypotheses, one reviewing and one checking the data. That kind of back-and-forth is a very different way to learn science.

But we cannot talk about the promise without talking about the risks, and there are several. The digital divide worries us most. Students at those universities can easily use the latest state-of-the-art AI infrastructure such as the AI Supercomputing lab and Digital Forensic lab, while rural and underfunded schools may fall even further behind. If this technology is meant to lift the whole country, national programmes need to prioritise lightweight, low-bandwidth AI models in Bangla that run on affordable hardware.

Then there is over-reliance. An agent can write code, draft hypotheses and run statistics by itself, and it is tempting for students to hand over their thinking too. Schools will need to stop grading only final answers and start looking at how students got there: their process, their questions and their judgement.

Data is another concern. For an agent to personalise learning, it needs a constant flow of information about each student. Without strong national protection, that opens the door to profiling and misuse, and it could erode trust in digital education altogether.

And finally, technology can support teachers, but it cannot replace them. Values, empathy and personal mentorship still come from people. The best use of AI agents is to take routine admin work off teachers’ hands, so that they have more time to nurture ethics, curiosity and resilience in their students. The potential here is hard to deny. If institutions lead on research while the country commits to equal access and ethical oversight, Bangladesh can build a scientific workforce that competes globally. In the end, the future of science here would not be decided by technology alone. It will depend on how wisely we choose to use it.
_________________________________
Dr. Carmen Z. Lamagna is Former Vice Chancellor and Member of the Board of Trustees, and Prof. Dr. Dip Nandi is Dean of Faculty of Science and Technology at American International University-Bangladesh (AIUB)