In what sense? It can be very useful as a source of knowledge to bounce off for CFD concepts It can be very useful for some case configuration, and for coding openfoam boundary conditions, function objects, solvers, etc It's very useful for writing code to post-process cases But if you describe a problem to it and ask it to make you an openfoam case, it will almost certainly get it wrong If you try to replace CFD altogether with a machine learning model that guesses a solution based on being trained on other cases, it will be wrong as soon as you ask it to consider anything that is not extremely similar to its training data