Turn-taking is the hidden interface
Fast answers matter less than knowing when to listen, wait, interrupt, and stop.
Read guide ↗Field 02 / Conversation design
Make the next turn clear.
Conversation design is not just tone. It is how people understand state, repair a misunderstanding, recognize evidence, and remain in control. Start with a single turn, then follow the commitments across the whole task.
The question to keep asking
What will the person believe happened after this response?
A useful review starts here
- Uncertainty prompts a useful next step.
- Corrections invalidate dependent information.
- Pending work is never described as completed.
- 01
Ask the useful question ↗
Separate plausible paths without turning chat into an interrogation.
- 02
Draw the state changes ↗
Specify guards, invalidation, correction and exits.
- 03
Tell the truth about waiting ↗
Keep accepted, running, completed and cancelled distinct.
See the source / related interface context
A canvas needs a place for uncertainty.
Keep deterministic steps visible, and make the generative parts inspectable.
Open the case file ↗This source capture illustrates a related interface—not evidence that the vendor passed these review checks.
The complete conversation design desk
Go deeper.
Clarify without interrogating
Ask the smallest question that safely separates the plausible paths.
Read guide ↗Write responses for action
A good bot answer is not an essay; it is a clear decision surface.
Read guide ↗Accessible conversation is more than screen-reader support
Design message pacing, focus, alternatives, error repair, and help as one accessible system.
Read guide ↗
NVIDIA · source visual / reviewConversation design3 minAuthoring interruptible voice turns: VAD versus semantic endpointing
Choose turn boundaries by task and failure cost, then make barge-in, partial speech, tools, and cancellation obey the same state machine.
Read guide ↗
Dify · source visual / reviewConversation design3 minVisual chatbot workflows: make state transitions inspectable
A practical way to draw branches, state changes, retries, and exits so a canvas remains an operating model rather than a decorative diagram.
Read guide ↗Long-conversation context budgets: allocate, do not just trim
A long conversation needs explicit space for current intent, pinned constraints, verified evidence, and a fallible summary—not a mystery pile of old turns.
Read guide ↗Accessible streaming responses: make partial output interruptible and readable
Incremental text can be useful, but it must not steal focus, flood announcements, or leave a broken transcript when someone stops it.
Read guide ↗Names and numbers in speech: repair identifiers before they become actions
Dictated identifiers need an explicit readback and correction path; recognition confidence is not evidence that a number or name is right.
Read guide ↗Keep the system connected
Foundations Handoff & service Put the reading to work