Context is the interface

The prompt box is the most visible part of an AI product, but it is not the whole interface. It is only the place where a request begins.

The deeper interface is the context around that request: the sources, assumptions, decisions, prior drafts, open questions, and constraints that should shape the answer.

A better prompt is not enough

Prompt craft matters. Clear language helps. Examples help. Constraints help.

But when the work becomes real, the issue is not only how to phrase the next prompt. The issue is how to preserve the material that should inform it.

Most useful work produces context:

• A decision that should constrain future drafts.
• A definition that should stay precise.
• A comparison that should be reused.
• A source that should support several claims.
• A question that should remain open until answered.

If those pieces stay trapped in a thread, each new prompt has to rebuild the environment from memory.

Prompt libraries try to solve this by saving reusable instructions. They help, but they are incomplete. A saved prompt can describe a workflow. It cannot by itself show which sources, decisions, drafts, and risks currently belong to this project.

Context needs shape

Context becomes useful when it has boundaries. A source should not be confused with a decision. A task should not be confused with a claim. A rough note should not be treated as a stable conclusion.

BaseHalf gives context shape through Maps and Points. The Map shows what exists. The Point gives one piece of work a place to be opened, edited, and reused.

This changes the interface. Instead of asking the model to infer the whole project from the latest message, you can show it which context matters.

Shape also makes context deletable. If everything is one transcript, removing a stale assumption is awkward. If the assumption is a Point or Block, it can be revised, archived, or disconnected without disturbing the rest of the work.

The next interface is compositional

Good context should be portable. It should be possible to apply one piece independently, combine it with another, and carry it into the next question.

That is why composability matters. A decision can guide a draft. A rubric can evaluate several ideas. A source can support more than one claim. A review task can point back to the concept it tests.

The interface is not only the message you type. It is the system that lets context be selected, connected, and reused.

Less magic, more control

There is a temptation to imagine AI products as invisible intelligence that remembers everything and simply knows what to do. That can feel magical, but it also removes control.

Visible context is different. You can see what the model is using. You can revise it. You can remove stale material. You can decide which parts should travel forward.

The best AI interface may look less like a blank chat and more like a working surface for context. The prompt still matters, but the context becomes the product.

What to design for

If context is the interface, product design should optimize for a few concrete abilities:

• create a useful piece of context quickly
• name it so it can be found later
• open it when it needs careful work
• connect it when it should inform another Point
• remove it when it becomes stale
• combine it with other context for a new question

The best interface is not the one that remembers everything automatically. It is the one that gives people control over what should keep working.