Most debates about cloud architecture focus on the wrong question. Teams argue over which database or identity provider is best for their workload. The better question is, how hard will this be to change in three years?
Teams treat certain foundational choices as technical decisions they can revisit later. It won’t be that simple. By the time you realize a choice has become permanent, changing it means rebuilding systems that already work.
Your AI embedding model looks like an implementation detail
When integrating Large Language Models (LLMs), developers often assume they can easily swap out the embedding model later if a cheaper or faster one becomes available.
Why it becomes permanent:
Embeddings are mathematical representations of your data. If you generate 5 million vectors using OpenAI’s text-embedding-ada-002, and later want to switch to a local open-source model, you cannot just swap the API endpoint. You have to re-process and re-embed all 5 million documents.
Teams treat certain foundational choices as technical decisions they can revisit later. It won’t be that simple.
