Error in Progress explores how to improve error recovery within voice-first multimodal AI assistants. The project investigates how users behave during misunderstandings, what expectations they hold, and how they attempt to repair communication when the system fails. These insights inform design strategies that blend voice, visual elements, and haptic feedback to provide clearer, more supportive responses. Through prototyping and user testing, the project demonstrates how multimodal interactions can lower frustration, enhance communication clarity, and restore user trust at critical moments. Rather than treating errors as breakdowns, the approach reframes them as opportunities for adaptive, empathetic interaction. The goal is to create AI experiences that are more resilient, transparent, and user-friendly, ensuring that everyday assistants can handle errors gracefully and maintain meaningful engagement even when things go wrong.