Microfluidic design optimization in the era of AI agents

Abstract

Artificial intelligence (AI) is increasingly used to optimize design of microfluidic devices by predicting their performance, reducing the number of simulations, and guiding the search for design configurations with improved performances. However, in most studies, the design problem is still defined by the researcher. The device concept, design variables, parameter ranges, objective functions, and constraints are selected in advance, while AI is used to search within these predefined limits. A key limitation of current approaches is that AI is involved only in a specific step of the workflow, performing a narrowly defined task. The emergence of AI agents may change this workflow more fundamentally. AI agents are autonomous systems capable of coordinating computational processes and adapting their behaviour to achieve a defined goal. In other words, in an agentic design workflow, an AI system could interpret an optimization goal, select computational tools, define or modify the design space, run and analyse numerical simulations, and adjust the optimization strategy based on the physical behaviour of the system. This perspective examines how AI agents could extend microfluidic design optimization beyond fixed, researcher-defined search spaces toward adaptive and physics-grounded workflows. We discuss a spectrum of agentic approaches, ranging from LLM-driven conceptual design to agent-supervised numerical optimization, together with evolving role of human supervision and technical challenges associated with reliable implementation. Moving from fixed optimization workflows towards agent-guided design could enable microfluidic systems to be explored more efficiently and may shift design optimization from simple parameter search towards a more adaptive and reasoning-based process.