faster inference would lead to “just-in-time” code generation, where instead of creating MCPs/ CLIs/ steering you can just ask the model to do X and it can create series of cascading code based workflows without forcing users to go through existing code paths all w/ negligible