I like this. Couple things to add:

Fast setup and revision are important but incomplete list of maintenance tasks; are metrics/logs predictably named and accessed? Can you perform manual experimentation without hard-to-configure client (ir: hit the server with a browser or run a cli)?

Also, "cycle time" or "revision time" are soo important, but I havent found a good way to do that with AI model development :( any tips here?