Definition
Conversation design is the discipline of designing AI agent interactions as structured conversations rather than simple prompt-response pairs. It draws from UX principles established before the AI wave — notably from “Conversations with Things” by Diana Deibel and Rebecca Evanhoe (2021) — and treats agent building as a human-computer interaction challenge, not purely a technical one.
Key Points
- Building conversational AI agents is primarily a conversation design challenge, not a technical one
- The field predates modern AI: “Conversations with Things” (Deibel & Evanhoe, 2021) was published before ChatGPT
- Conversations are messy, cultural, and deeply human — agents must handle this reality
- Core elements: turn-taking, recovery/repair, personality, intent handling, context management
- The book opened the author’s eyes to the gap between what developers build and what conversation requires
Related Concepts
- turn-taking — When the AI should speak vs. wait
- graceful-recovery — How the agent handles failures
- personality-design — The agent’s consistent behavioral identity
- intent-classification — What the user wants to accomplish
- accommodation — How users adapt to the agent’s communication style
- agent-loop — Conversation design governs what goes inside the agent loop at each iteration
Implications
This matters because it explains why technically sound agents often feel broken in practice. The gap between “correct responses” and “good conversation” is where most agent failures live. Teams that invest in conversation design skills (linguistics, UX, cultural awareness) produce better agents than teams that only invest in better models or frameworks.
Open Questions
- Is conversation design a skill that AI engineers can learn, or does it require dedicated conversation designers?
- How do you test conversation quality at scale? Are there automated metrics for conversational UX?
- How will conversation design evolve as agents become more capable and handle more complex interactions?