Communication between humans is often a multimodal conversation. It can be a recurring loop with optional steps depending on the complexity of the situation and our possibly evolving understanding of what we want. The same is true of our communication with computers. When a human is reasoning about complex data and asking for automated assistance, this conversation can be especially hard. Carefully designed data rendering and interface affordances can help significantly. I will present a selection of novel interactive systems whose interface designs help users (1) quickly and accurately form a mental model of a concept from relevant data, (2) form - and update - their beliefs about what they want (3) communicate what they want to a computer when its automation capabilities are helpful and/or (4) have justified confidence that the automation being carried out reflects their current intentions. Systems discussed will include an interface for comparing and contrasting similar but distinct programming libraries based on real-world usage and a mixed-initiative neurosymbolic system that helps users perform qualitative coding, i.e., label large volumes of short texts with emergent themes.