Building a personal agent that manages texts and emails with persistent memory — but never sends without approval
A user sketches the ideal personal assistant: it reads texts and emails, remembers conversation history and relationships, pulls context from calendar and other sources, and drafts personalized replies for approval before sending. The discussion highlights two design rules that matter more than model choice — approval before any send, and memory stored as short notes per person rather than raw chat dumps — plus starting with one channel before adding a second.