Many AI users unknowingly craft prompts that lead to inefficient or biased outputs due to subtle wording issues.
A web app could ingest prompts and score them on clarity, potential bias, and likely computational cost based on known model behaviors.
Sell licenses to teams deploying AI at scale who need to optimize prompt pipelines.
Start with a simple scoring algorithm for popular LLMs and expand based on user feedback.
The biggest risk is prompt engineering becoming less relevant as models improve at understanding natural language.