The Problem of Engineering-Led Design

Modern AI applications are suffering from a branding and UX crisis. Companies like Google, Anthropic, and OpenAI are exposing their internal engineering architecture directly to end-users by forcing them to manually switch between different "modes" or "features" (e.g., Gemini’s Spark vs. Daily Brief, or Claude’s Chat vs. Cowork). This approach is fundamentally flawed because it requires users to understand the underlying capabilities of the AI rather than simply stating their intent. When users have to decide which "mode" to use, the AI ceases to be a helpful assistant and becomes a tool that requires manual configuration.

The Friction of Fragmented Interfaces

Google’s Gemini app serves as a prime example of this friction. By branding specific capabilities like "Daily Brief" as standalone features with their own navigation, Google creates unnecessary cognitive load. Furthermore, these features often fail to provide genuine value; for instance, proactive updates often surface irrelevant or "creepy" historical search data rather than actionable information. This suggests that these features are designed by engineers prioritizing technical capability over user utility. The industry-wide trend of separating "chat" from "work" modes—often without shared memory between them—further alienates users by breaking the natural flow of conversation and task execution.

The Case for Invisible AI

The most successful AI integrations will likely be those that remain invisible, enhancing existing workflows rather than demanding new ones. Apple’s approach to Siri is highlighted as a superior model: by making existing system features (Spotlight, Photos, Camera) smarter without requiring users to learn a new interface or switch modes, Apple avoids the "branding problem" entirely. Similarly, the rise of text-based AI assistants—which treat the AI as a contact in a messaging app—succeeds because it leverages the most intuitive, well-understood interface available: text messaging. The goal for AI product builders should be to minimize the distance between a user's intent and the AI's action, rather than forcing the user to navigate a menu of internal product names.