Give conversational interfaces a way back
2 min read
A conversation makes it easy to begin an ambiguous task. It becomes harder when the user must revise one detail, stop an action, or identify which answer is current after several retries.
Define those controls before adding more assistant personality. For a practical example of keeping direct manipulation alongside AI, see the editable output decisions from KIMEO.
Keep the conversation connected to the work
Show the document, task, or record the assistant is working with. Give users a way to inspect what context is included.
Allow structured controls where precision matters. A date picker or selection list can be easier to use than another conversational turn.
Let people revise an earlier choice
Offer direct editing of important inputs and make the effect of a revision clear. Preserve earlier output when it helps comparison.
Keep cancellation and retry understandable. The person should know which operation is still running and which result is current.
Write recovery messages as instructions
Describe what prevented progress and give a next step the user can take. Avoid vague failure messages that simply invite the same attempt again.
Test recovery with someone unfamiliar with the system. If the message needs an explanation from the team, rewrite it.
Put this into practice
Explore the related work and delivery scope.
- Related project: KIMEO
Inspect the published product workflow and delivery context.
- Explore the relevant service
See the engagement scope and start a specific project conversation.
Make conversation and direct control work together
Share the important user choices, external actions, and recovery cases in your assistant.
Discuss a conversational product
