PRODUCTION SCENARIO
A Generative AI Engineer redeploys a booking agent every time a new system prompt ships, which takes 40 minutes and blocks fast rollback. The prompts are already versioned in the MLflow Prompt Registry.
How should the deployed agent reference the prompt to remove that redeploy?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Load the prompt by an alias the team can repoint to a new version
A deployed app can load a prompt by an alias such as production, so shipping a new prompt means reassigning that alias to a newer version without changing or redeploying application code, and the earlier version stays available for rollback. Loading the newest version at startup still needs a restart, baking the text into the agent still needs a new model version, and reading a Unity Catalog volume file on every request adds a read to the hot path and gives up the registry's versioning and lineage.