Conversation Starters is a series of interactive prototypes that probe how to design explainable interactions with AI in everyday life. Taking a more-than-human approach, we explore how 'failures' could be transformed into opportunities for situated understandings of AI. We describe the process of designing fictional artifacts and scenarios about conversational agents that can grow at home. While overall the project suggests that misunderstandings could help people develop sensitivities for knowing when to trust AI systems, the metaphor of 'growing an AI' (which positions training as a matter of care), highlights that practices of sharing and experimenting could be valuable starting points for designing explainable and trustworthy interactions with of AI.