LLM Team Personality Impacts Performance Based on Task Structure.

Researchers investigated the influence of personality traits, assigned through prompting, on the performance of multi-agent Large Language Model (LLM) teams. Previous studies showed that LLMs prompted with low agreeableness tend to produce adversarial language, while high agreeableness leads to cooperative communication. This work systematically examined how these communication shifts translate to objective task performance across different domains. The findings reveal that the relevance of personality composition is highly dependent on the task's inherent structure. For highly structured tasks like coding, manipulating personality traits, such as low agreeableness, caused significant changes in communication style but had minimal impact on task completion. Conversely, in less structured environments like open-ended research collaboration or competitive bargaining, the same personality manipulation substantially degraded the LLM team's overall performance. This suggests that designers of multi-agent systems need to consider task type when applying personality prompting.