Thanks for the response. Good point about data representation in COBOL being optimized for financial calculations. I think that the data representation issues could be handled in a post processing step, but I don't think that is even the critical problem. From the experiments I have done so far, it's hard to get Codex to produce Python output that resembles working code. JavaScript is a more promising target (see https://towardsdatascience.com/codex-translates-cobol-into-javascript-266205d606c0 ), but I don't know if JavaScript is a credible language for the kind of bedrock applications that are still running in COBOL, and if JavaScript isn't the right landing spot, how hard it would be to hop to another modern language, like C#.