Should AI Prompt Stores Control Model Selection? A .NET Developer's Take

A developer analysis questions whether tools like Langfuse should bundle model configuration—such as model name, token limits, and temperature—directly alongside prompt definitions in .NET applications using Microsoft.Extensions.AI. While the approach enables non-engineers to update prompts and model settings without redeployment, critics argue it mixes low-risk content decisions with high-stakes infrastructure choices in a single untyped JSON blob. This coupling means a typo or misconfiguration in the prompt store could silently cause failures or unintended behavior only when live traffic hits the system. Managing environment-specific model settings, such as using cheaper models in development versus production, also becomes cumbersome under this design. The post proposes separating model configuration into typed, validated application settings so errors surface at startup rather than at runtime.