Home / Work / Chat Bar

The seams of a conversation: Chat Bar

Message design, latency states, and what happens between the user's intent and the system's answer.

Chat Bar — AI
Role
Product Designer
Timeline
2023
Team
Product, engineering
The challenge

Chat looks like the simplest interface in the world and is one of the hardest.

A blank input tells the user nothing about what the system can do. Latency is visible in a way it is not elsewhere. Long answers become walls that cannot be scanned.

Opportunity

The interface lives in the gaps

Entry, waiting and formatting decide whether a conversational product feels capable or vague.

The solution

Work on the seams rather than the bubbles.

Entry points that suggest capability without prescribing it, waiting states that make latency legible, and formatting that gives long responses structure.

Design decision

Capability without a menu

Suggest what is possible without limiting it.

Entry points hint at range rather than listing commands, so discovery happens without turning an open interface into a form.

Design decision

Legible waiting

Streaming and pending states designed, not defaulted.

Waiting is understood rather than endured when the interface shows that something is happening and roughly what stage it is at.

Design decision

Correction inside the message

Retry and feedback where the content is.

Fixing a bad answer happens at the answer rather than in a menu, which is the only place a user thinks to look.

Outcome

A surface where capability is discoverable.

The patterns generalise to any model-backed conversational product, which is why they carried into later AI work.