Fluorescent protein-based biosensors are indispensable molecular tools in cell biology and biomedical research, providing non-invasive dynamic measurements of metabolite concentrations and other cellular signals. Traditional methods for developing these biosensors rely on rational design, but directed evolution methods offer a more efficient alternative. This review discusses recent advancements in the development of biosensors using directed evolution, including methods for optimizing domain fusions, sequence optimization, and new screening and selection systems. Additionally, the incorporation of machine learning into the directed evolution process is explored, highlighting its potential to enhance the efficiency and cost reduction of biosensor development. Finally, emerging trends in the development of near-infrared biosensors and photochromic sensors are discussed, along with the opportunities presented by de novo design of sensing domains and biosensors.