Intent Slot

An intent slot is a named typed region inside a generative interface that tells the model what content and components fit a specific part of the user request. It exists because raw models treat every prompt like an empty artboard and spit out layouts that feel random. Designers stopped shipping fixed screens and started shipping these slots the same way they once shipped grid systems. The slot acts like a brief for a junior designer who happens to be a frontier model.

Each slot carries rules about allowed components and forbidden ones. A primary answer slot might take a callout or chart but never a full form. A suggested next step slot might accept one CTA card but never three paragraphs of explanation. You feed these rules in the system prompt and the frontend renders them in a stable grid so the product still feels like one product.

Do not mistake intent slots for static wireframe sections. Static sections never change their job or their layout. An intent slot keeps its position and purpose while the exact component inside it shifts with every new user question. It is also not a loose container div you label after the fact. Slots are the first design artifact you create and the one the model is graded against.

Teams that skip slots regret it immediately. The generated output turns into a choose your own adventure of inconsistent layouts. Users see forms where summaries should be and data tables where actions belong. What looks like innovation in a demo becomes support ticket volume in production. The model is not guessing your brand standards on its own.

Concrete example shows up in the Vercel AI SDK patterns that shipped in early 2025. Teams defined three slots called primary result evidence and next action. The model streamed only allowed components into each one. The frontend kept a fixed two column grid around them. Every report looked like it came from the same design system even when the underlying data and layout logic changed completely between users.

Claude Artifacts in late 2024 used the canvas as a persistent intent slot for the main artifact while the chat handled intent collection. Users could regenerate or edit in place without breaking the surrounding controls. The whole experience stayed coherent because the slots and their recoverability affordances were designed before any code was written. Bolt took the same idea further by exposing starter prompts that mapped directly to slot combinations.

Use intent slots on surfaces where the same question from two users should produce different content but the same brand feel and interaction quality. Personalized dashboards internal analytics summaries and dynamic help pages earn their keep here. Do not use them on checkout flows login sequences or any regulated surface where layout deviation equals compliance failure. The tradeoff is more system design work upfront against far less per surface design work later.

The teams winning in 2026 treat slots like the new wireframes. They version them review them and run them through the same evals they run on components. A slot without clear rules is just marketing copy for randomness. A slot with tight contracts is how you scale design quality faster than the models improve.

If your designer cannot list the slots and their acceptance criteria from memory the vocabulary does not exist yet. The model is simply guessing in public and hoping the brand team does not notice. That works in demos. It fails in shipping products.

Intent slots force the model to stop improvising and start composing. That single change separates interfaces that feel designed from interfaces that feel generated.