Would it make sense for AI services to have a memory that can be used across applications (coding, meetings, planning) without having to build a product that does all?