The value right now is learning which prompts work, which strategies work, what is the sweet spot in terms of model size, is it better to run a model that costs

The value right now is learning which prompts work, which strategies work, what is the sweet spot in terms of model size, is it better to run a model that costs 10 times more per token vs smaller inferior models that might need have 3 attempts at it, what are the best RAG and other enrichment strategies etc. You can only really get that when you have a large market share. When your IDE makes a call to OpenAI they get to see only half the picture they don’t know how the user reacts to the output. So there is a lot of value in being in the client loop also and seeing the full picture.