Experience design for ai agents : an analysis of human-centered guidelines in llm-owning companies

Abstract

Recent advances in large language models have driven the expansion of artificial intelligence agents, capable of planning and executing actions autonomously to achieve goals, shifting the challenge from technical development to the design of human-centered experiences. In this scenario, the aim of this research was to analyze how LLM-owning companies structure and communicate design principles for AI agents. Through a thematic documentary analysis of public materials from 22 companies, only three organizations (OpenAI, Google, and Microsoft) were identified as providing consolidated guidelines on agent design. The comparative analysis revealed convergences and gaps between these corporate principles and academic frameworks of human-AI interaction. The main results are: (i) the identification of five thematic axes that structure theoretical-practical alignment, AI visibility, contextualized interactions, feedback mechanisms, adaptive evolution, and trust; and (ii) the mapping of significant asymmetries, such as the absence of prototyping guidelines and iterative error treatment in some companies. It is concluded that, although there is progress toward human-centered practices, corporate guidelines remain fragmented and lack integration with consolidated principles from the literature, indicating opportunities to improve the design of agents that are more ethical, understandable, and aligned with human agency.