Hwo to deal with agents versioning

[pedrorjbr]

I am building an AI workflow using langgraph. This workflow could take hours to day to complete, and it has state. Suppose that I have made a change on this workflow ( change the prompt or even change the graph architecture), and would like to deploy it. The workflows that are already running could not deal with this new versions, as it's already in "running" state with an old version.
I suppose that I need to implement some kind of versioning. And I suspext that it's a common use case. How are guys dealing with?