What is a conversational UI?
A conversational UI is an interface where people use software through a back-and-forth exchange, typed in a chat or spoken aloud, rather than by clicking through screens. It keeps the context of the exchange, so each turn builds on the one before.
What is a conversational interface?
A conversational interface is the same idea as a conversational UI, and the two terms are used interchangeably: the user and the software take turns in a dialogue, and the dialogue is how work gets done.
What makes it conversational, rather than just a text box, is memory of the exchange. A search bar treats every query as new. A conversational UI understands that "and the one from March?" refers to the invoice you just discussed, and that "send it to finance" means the report it just built.
Conversational UIs show up in two forms:
- Chat. A message thread, usually in a side panel or a full screen. Good for precise requests, pasting in details and reviewing results.
- Voice. Spoken requests and spoken replies. Good when hands or eyes are busy, on mobile, or when typing the request would take longer than saying it.
Many products now offer both in the same panel, so a user can switch between typing and speaking.
What are the types of conversational UI?
The type depends on what sits behind the conversation.
| Type | How it works | Where it falls short |
|---|---|---|
| Scripted | Fixed decision trees with buttons and set phrases | Breaks when the user goes off script |
| Answer-only | A language model answers from docs and help content | Explains the steps, then leaves the user to do them |
| Agent | A language model that also calls the product's functions | Needs permissions, confirmations and clear limits |
Scripted flows still work for narrow jobs like booking a slot or collecting a few fields. Most of the recent growth is in the other two types, and the line between them is the one that matters most.
Should a conversational UI answer or act?
Answering versus acting is the practical split in conversational UI today. A chatbot answers. An agent acts.
Say a user on a billing platform types "why did the Acme invoice fail?" An answer-only interface finds the help article on failed payments and explains how to update a card and retry. The user still has to find the invoice, open the payment settings and click Retry.
An agent reads the invoice, sees the card expired, says so, and offers to send Acme a request for a new card or retry with the backup card on file. The user picks one and confirms. The conversation ends with the work done, not a set of instructions.
Answering is enough for "what does this setting do?" Acting matters for "set it up for me." Most products need both, and the design work is in deciding which actions the agent can take, which need the user's confirmation, and how the user sees what happened.
Foldspace builds the acting kind: the Product Agent answers, acts and navigates in the product, by chat or voice, calling the APIs the product already runs on, with the signed-in user's own permissions.