Source: arXiv — 2026-07-06
Summary
"Agents with Feelings? Personality and Emotion in Multi-Agent Software Teams" studies how assigning psychological trait profiles to LLM agents affects multi-agent software engineering team performance. Using a framework combining Big Five personality traits, basic emotions, software-engineering-relevant work styles, and task roles, the authors benchmark how different team compositions perform on code generation and code review — treating agent "personality" as a controllable design variable rather than an incidental prompting detail.
Key Takeaways
- Tested 78 team-profile configurations across 4 LLMs and 659 task instances.
- The gap between the best and worst shared-profile (homogeneous-personality) team configuration was 7.1–11.3 percentage points in pass@1 on code generation.
- Mixed-profile (heterogeneous personality) teams beat the best homogeneous-profile team in 6 of 8 model-task settings tested.
- Authored by Yunyan Ding, Thomas Zimmermann, and Iftekhar Ahmed.