In this study, our central aim is to enhance Computational Fluid Dynamics (CFD) simulations by integrating Artificial Intelligence (AI), with a specific focus on approximating predicted fields to converged steady-state solutions. We propose a workflow leveraging a Neural Network (NN) as a predictive model, utilizing an incremental training paradigm to address the substantial temporal expenses associated with dataset creation. Our investigation revolves around 2D CFD cases, particularly within a water tank configuration featuring a momentum source to induce fluid mixing. Through iterative application of our methodology, significant acceleration of 2D simulations is achieved, resulting in efficiency gains of approximately two-fold. Furthermore, the iterative process enables the accumulation of an expanded dataset, fostering potential for further acceleration and scalability of the AI-driven workflow. Our findings reveal a cumulative speed-up of 2x, corresponding to an estimated reduction in computation time of 73% in the build of the entire dataset.