Jannik Endress
Graduate Visiting Researcher Student, Graduate School of Business
Bio
Jannik Endress is a Visiting Student Researcher at the Stanford Graduate School of Business. He studies how populations of large language model agents form shared conventions and coordinate under different network structures, with Douglas Guilbeault in Organizational Behavior.
He is completing an M.Sc. in Management and Technology (Computer Engineering) at the Technical University of Munich. Before Stanford he worked in early-stage venture and consulting in Germany, including roles at Alstin Capital, Exxeta Ventures, and product internships in AI-enabled operations.
Research Interests
-
Artificial intelligence (AI)
Current Research and Scholarly Interests
I research emergent behavior in multi-agent LLM systems, with an emphasis on coordination, convention formation, and collective stability.
As models are deployed in interacting groups rather than as single assistants, new failure modes appear at the population level: runaway consensus, brittle lock-in, and coordinated drift away from intended behavior. These are AI-safety questions about what collectives do, not only about what one model says.
My current work treats naming-game experiments as a controlled setting for that problem. Populations of LLM agents interact on structured networks and must converge on shared labels. I vary topology and introduce committed-minority perturbation to see when conventions emerge, when they stabilize, and when a small set of stubborn agents can capture or collapse the group. Human coordination experiments (Ashery et al., Science Advances, 2025) are the benchmark.