Every AI (AWS, Google, and Genesys) have their own good and bad points. I feel trying to build full self-service in these bots is probably harder than it is worth, especially if you don't work in them as much I have. I have come to realization that DialogFlow, Lex and DialogEngine have unique functions and if you keep them specific to the task of NLU intent return, then you will have an easier time creating bots and have expectations met. For instance, if you use the bots within you flow to just gather intent, then you have very simple bots on these platforms that can perform that function and then you can tune and train them for that purpose. If you start having your bot in these platforms talk to back end systems, keeping track of variables, and start processing information, they can become very complicated very quickly.
A better approach is to keep you flow in Architect for all processing, decisions, and tracking and just use these bots for alphanumeric input or determining intent. Now that Genesys Cloud had DialogEngine, there is really no need to go out to Lex or DialogFlow for simple NLU functionality and the tools provided by Genesys are far better than AWS or Google when it comes to training and intent management.