Designing LLM-first products, not just features

The emergence of products with large language model (LLM)-powered features has introduced new possibilities in user experience design, moving beyond traditional integrations like chatbots or summarization tools. While many current LLM applications are merely add-ons to existing systems, the concept of LLM-first design proposes a radical shift, where generative capabilities become central to the user experience, transforming LLMs from passive assistants into autonomous actors capable of decision-making and complex tasks. This approach impacts both front-end and back-end development, encouraging conversational interfaces and sophisticated data architectures that enhance interaction and adaptability. Such systems challenge traditional testing and maintenance methods by introducing dynamic behaviors and user experiences. A case study of an expense reimbursement app illustrates the evolution from feature-based LLM integration to an agent and platform phase, where natural language becomes the primary interface for task execution. Ultimately, LLM-first design reframes the relationship between humans and software, shifting from static, form-driven workflows to adaptive, language-centered systems that promote a new simplicity in product interaction.