This paper introduces a new approach to on-line model-based optimization. The proposed approach differs from classical real-time optimization in that waiting for steady state is not necessary and novel optimization concepts are used in the solution procedure. Instead, current plant set points are periodically improved following periodic updates to the model. Thus, realtime evolution (RTE) is based on the continuous improvement of plant operation rather than on the optimization of a hypothetical future steady-state operation. Despite using a simpler scheme, the proposed strategy offers a faster response to disturbances, better adaptation to changing conditions, and smoother plant operation, regardless the complexity of the control layer. In this paper, RTE design principles are conceptually presented and illustrated with relevant examples. Case studies are considered to validate the proposed methodology and to discuss the results obtained. First, a classical CSTR benchmark is contemplated. Then, a real scenario is considered (a pilot plant at the UPC) where the RTE approach is currently implemented.