One Is Not Enough: How People Use Multiple AI Models in Everyday Life
New research finds users create shifting hierarchies and personalized switching patterns between competing AI systems.
A new study accepted to the CHI 2026 conference, titled 'One Is Not Enough: How People Use Multiple AI Models in Everyday Life', reveals that power users are not monogamous with their AI assistants. Conducted by researchers from KAIST and Seoul National University, the work involved a diary study and semi-structured interviews with 10 participants. It found that individuals actively use multiple competing systems—such as OpenAI's GPT-4, Anthropic's Claude, and Google's Gemini—concurrently, selecting each based on perceived strengths for specific tasks. This creates a complex, user-managed ecosystem far beyond single-agent interactions.
This cross-platform practice introduces significant coordination challenges that current tools don't address. Users must adapt prompts to different interfaces, calibrate trust against inconsistent model behaviors, and juggle separate conversation histories. The research identifies that users construct fluid hierarchies, designating primary and secondary models that shift with context. They also develop sophisticated, personalized switching patterns, often triggered by the need to aggregate tasks, adjust for effort and latency, or verify output credibility. These findings expose a critical gap in Human-Computer Interaction (HCI) design, which has historically focused on single-model interactions, and point to a major opportunity for new tools that support multi-MLLM workflow orchestration.
- Users create shifting hierarchies, designating primary and secondary models (e.g., GPT-4 for coding, Claude for writing) that change based on context.
- Personalized switching patterns are triggered by needs like task aggregation, latency adjustment, and output credibility checks, creating a user-managed workflow.
- The study highlights a major HCI design gap, as current tools fail to support the coordination challenges of using multiple AI systems simultaneously.
Why It Matters
This signals a shift towards AI orchestration as a core user skill, demanding better tools for managing multi-model workflows in professional settings.