According to a new preprint on arXiv, people are increasingly using multiple multimodal large language models (MLLMs) at the same time, rather than relying on a single tool. The study, titled "One Is Not Enough," suggests that users pick each model based on what they see as its particular strengths, leading to a workflow that spans several platforms.
This cross-platform habit comes with its own difficulties. The abstract notes that users face coordination challenges, including the need to adapt their prompts to different models. The full details of these challenges are not yet available in the abstract, but the finding points to a shift in how AI tools are integrated into everyday tasks.
The research adds to a growing picture of AI usage as a multi-tool practice, where no single model dominates. As the practice becomes more common, the friction of switching between models may become an important design consideration for AI developers.