Core Surface

Core surface is the principle that positions the AI model as the first surface a user encounters when they open the product. The prompt field or the model generated view claims the central real estate and the primary workflow. The rest of the UI becomes scaffolding built to feed the model good context, display its outputs, offer transparency into its decisions, and provide clear undo paths when it acts with agency. This is the foundational posture for any AI native product. It shows up in the deletion test as the difference between a hollow shell and a fully functional legacy app. Cursor fails the deletion test in the best way possible because its entire value proposition collapses without the model. Perplexity has no reason to exist without the model at the center of every search. The principle connects directly to the other five. Prompt as input only makes sense if the prompt lives on the core surface. Agency by default requires that the core surface show what the model did. Transparency surfaces belong inside the core view not in a hidden panel. Deliberate model reveal keeps the core surface clean for normal users. Latency as a design constraint matters most when the core surface is the only surface. Teams that adopt this posture from day one ship different products than teams that add it in sprint twelve.

Core surface is not a side panel with a sparkle icon. It is not the chat window docked to the right of a CRM dashboard that users open once and then ignore forever. It is not the pattern that dominated enterprise SaaS in 2024 where teams kept their old UI intact and bolted on an AI feature that could be ignored without losing any core functionality. Those designs fail every real test. The cold open test shows the old dashboard first. The deletion test leaves a working product behind. The onboarding test shows users completing tasks the old way. It is not the sparkle button added to every text field that creates barnacles of UI noise across the product. It is not early versions of Notion AI that lived behind commands inside documents that worked fine without them. It is not Cluely acting as a full screen overlay while the underlying app stays completely untouched and primary. Those are all side panel postures dressed up as innovation. They produce low engagement metrics and blog posts that quietly stop mentioning the AI feature six months after launch.

Perplexity ships core surface better than almost anything else in consumer tools. The product opens to a prominent prompt bar and the entire experience flows from there. The model is the search engine, the browser, the summarizer, and the answer surface in one. Citations appear inline as part of the core view so users verify claims without leaving the main flow. Streaming responses keep latency feeling alive. Arc Search does the same for the mobile browser category. Instead of returning a list of links the model replaces the tab with a synthesized answer surface complete with source links at the bottom. Users tap one button and the model does the browsing, reading, and distilling. Cursor applies core surface to the entire code editor. Cmd K is not a side panel. It is the primary way to interact with the codebase. Agent mode acts with high agency across files then presents diffs as the transparency and recovery surface. Model selection stays hidden until power users seek it. Granola makes the model the core of its value even though the input starts manual. The augmentation layer that runs on the notes turns the model into the reason users return to the app. The raw transcript sits next to the augmented version as a built in transparency surface. Linear takes a quieter route that still qualifies. The model lives inside the command bar that serious users open constantly. Natural language commands become prompt as input. Issues get created with agency and appear in the timeline for transparency. Krea turns the prompt plus reference images into the central canvas that drives generation. Lovable starts every session with a single prompt surface that builds entire applications. Each of these products made the model the load bearing wall. The cautionary tales of 2024 side panel products show what happens when teams refuse this rebuild.

Use core surface when your product value dies without the model and you are ready to redesign the primary workflow around it. Deploy it on new consumer apps like Perplexity and Arc Search where cold open leads directly to a prompt. Apply it to professional tools like Cursor and Linear where power users live in the surface that the model now enhances. Bring it to ambient tools like Granola where the model augmentation becomes the central value. Combine it with the full set of principles for best results. Test it with the pre ship checklist. The cold open must hit the model surface. The deletion test must leave a hollow shell. The form to prompt audit must show replacement not coexistence. Avoid core surface when retrofitting rigid enterprise systems that cannot replace forms with prompts due to compliance or when the main use case is structured data entry that models still handle poorly in 2025. Do not use it if your team is not ready to own the latency rhythm or build proper transparency surfaces. The sparkle button audit should come back with fewer than three icons or you are decorating instead of redesigning. Most teams in 2024 failed these tests and shipped AI that nobody used.

Core surface decides if users open your product to use AI or if they can ignore the AI and still get their work done.