Interfaces beyond the chat box

Words and state

Conversation is well suited to expressing intent. A person can describe a goal, add a constraint, or ask for a different approach. It is less obvious that a transcript is the best place to represent the state of ongoing work. Which draft is current? Which actions are waiting? What changed since yesterday? Important answers can become scattered across a long sequence of messages.

Traditional interfaces solve some of these problems by putting state in a visible place. A calendar represents time; a board represents work; a map represents location. Those structures narrow the possible actions, but they also reduce the amount a person has to remember. Intelligent products could combine that clarity with the expressive flexibility of language.

Consider an assistant helping plan a trip. A conversation may be the right way to understand preferences. A map may be the right way to compare neighborhoods. An itinerary may be the right way to inspect timing. The interface would change shape around the task while preserving a shared understanding of what the person is trying to do.

Give the work a place

One hypothesis is that useful intelligent interfaces will give their results a durable place. A design lives on a canvas. A comparison lives in a table. A proposed code change lives in a diff. Conversation can remain nearby, helping the person explain what to adjust, but the work itself should be inspectable in the form that best reveals its quality.

This raises an interesting product question: which parts of an interface should be generated, and which should stay predictable? A system that rearranges every control could be difficult to learn. A system that never adapts could miss the opportunity to present a better working surface. The right balance might be stable foundations with flexible representations of the current task.

Consistency would remain valuable. A person should know where to find their work, how to undo a change, and what a button will do. Adaptation should reduce effort rather than turn the product into a puzzle. An intelligent interface can still have a calm, recognizable grammar: here is the result, here is the evidence, here is the next decision.

The cost of explaining

The empty prompt also shifts responsibility to the user. A person may need to know what the system can do before asking for it. They may need to repeat context that already exists elsewhere. They may need to invent a vocabulary for an operation that a direct control could express in one click. Language is flexible, but flexibility does not always mean ease.

Imagine selecting a paragraph and seeing a few relevant options: shorten, clarify, compare with the previous version. A short instruction could still handle an unusual request. The visible actions would make common possibilities discoverable. This is a small example of an interface that meets the person halfway instead of requiring them to compose a perfect request.

There is also room for quieter interaction. A system might prepare a proposed change and place it where the user is already working. It might surface a missing assumption alongside a chart. It might offer a useful next step when a task reaches a natural pause. Each possibility needs care: the same suggestion can be helpful in one context and distracting in another.

Start with the task

For a founder exploring this space, the first question may be about the work rather than the interface. What does the person need to understand? Which decisions are consequential? What information should remain visible? Where does the current product ask the user to explain something it could already know? A useful shape can emerge from those answers.

None of this makes conversation a dead end. It suggests that conversation could become one part of a richer experience. Some products may be best as a simple exchange of words. Others may need a workspace, a timeline, or a new structure entirely. The open question is how to preserve the ease of expressing intent while making the resulting work tangible enough to trust.