working on a project slightly long (@spently) is now making me realise the hard truths of scaling AI apps
- prompt optimization and importance of evals
- continuity when models are decommissioned/dropped
It all works at start, but maintaining it is a good engineering problem
