AI for the data, Science for the insight: a working model for scientific research

Abstract

Integrating artificial intelligence (AI) into science requires defining which tasks to automate and which demand human judgment. This article proposes transforming workflows in three steps: mapping dependencies, scripting with AI assistance, and using orchestrating agents. This methodology accelerates data processing up to 80 times and applies equally to university administrative burdens. A key distinction is emphasized: scripts automate and agents orchestrate, but neither replaces indispensable scientific judgment. Delegating critical data interpretation to AI often yields conclusions lacking physico-chemical rigor. The study concludes by analyzing the implications of these tools for academic training, university-industry collaboration, and the current limits of technological autonomy in experimental research.