AI in Scientific Research: From Chatbots to Agentic Workflows
Important
information
AI in Scientific Research: From Chatbots to Agentic Workflows
In the past year, AI in research workflows has moved beyond chatbots that answer questions. AI agents can now work with your files, run analyses and produce results you can check. This brings new opportunities for research, along with new questions about reliability, ethics and good scientific practice. This workshop gives researchers an up-to-date overview of the field and practical experience with agentic workflows.
Part 1 – Seminar
Effective and Safe Use of AI in Scientific Research.
A theoretical session on how AI is used in research today. We will cover the latest developments, practical benefits, risks and AI ethics, plus current journal policies on AI use. We will also discuss strategies to prevent “deskilling”, so that AI strengthens your scientific expertise as you bring it into your workflow rather than replacing it.
Part 2 – Hands-on Workshop
Building AI Agent Workflows.
A practical session on AI coding agents, which are proving to be very powerful tools for researchers. Unlike a chatbot, an agent works directly with files and folders on your computer. It writes and runs scripts for literature and data analysis and leaves outputs you can check. Participants will build their own agentic workflow to help with their research tasks.
About the trainer
Michaela Liegertová. Vice-Dean, Faculty of Science. UJEP, since 2018. “I see AI in science from three sides at once — using the tools in my own research, helping write the institutional rules as a vice-dean, and listening to the national policy conversation. Those three worlds barely talk to each other, and connecting them is a lot of what I care about. The bench work underneath it is extracellular vesicles, genomics, and metabolic and reproductive disease. That’s the ground I stand on, and it’s why I don’t trust a result just because a model produced it — in biomedicine a hypothesis isn’t enough; you need mechanism, controls, and eventually wet-lab validation. A short version of my whole approach: generation is no longer the bottleneck — validation is.”