While building AI agents with customers at Famulor, we noticed another recurring problem:
As soon as conversations become non-linear, prompts break down.
People try to:
Stuff complex logic into one giant prompt
Explain branching behavior in long paragraphs
Maintain if / else logic inside plain text