AI
Designing a better human handoff
A sample exploration of where an AI assistant should step back, and how a thoughtful handoff can keep a conversation moving.
This is sample content, written to demonstrate the blog layout. Replace it with your own article before publishing your journal.
An assistant can answer a straightforward question in seconds. The more interesting design problem begins when the conversation stops being straightforward.
Start with the exception
Imagine a client asking about a booking, then mentioning a concern that needs someone who knows their history. Continuing down a scripted path would miss the point of the conversation.
A useful workflow has a clear exit. It recognises when the next step belongs to a person and explains what will happen next.
A good handoff preserves context, so the client does not have to start again.
Give the next person enough context
The handoff can be small. What matters is that it includes the information someone needs to continue:
- What the client is trying to do.
- What the assistant has already answered.
- The specific question that still needs attention.
- Whether the client is waiting for a reply.
The goal is a useful summary, not a transcript that takes longer to read than the conversation itself.
Make the waiting state clear
An acknowledgement should describe the next step in plain language. Avoid promising an immediate reply unless the team can actually deliver one.
For example: “I've passed your question to the team with the booking details you've shared. They'll continue the conversation here.”
That small piece of interface writing can make the transition feel considered.
Test the awkward conversations
- Try a question the assistant cannot answer.
- Change the subject halfway through a booking.
- Ask to speak with a person directly.
- Check that the team receives the context each time.
These are illustrative ideas, not results from a client project. A real article could include a workflow diagram, observations from testing, and the changes made along the way.
For a related sample, read Start with one repeatable task.