Project Hollow
Choir
Coordination without instruction.
Overview
Multi-agent systems research studying how LLM agents develop emergent roles, division of labor, and consensus mechanisms without being explicitly assigned instructions.
Research Focus
The Research Problem
When scaling multi-agent networks, rigidly pre-assigning roles causes brittle task bottlenecks, while leaving agents completely unguided leads to duplicate execution and high token waste. Understanding how coordination patterns form naturally in shared environments is critical for building resilient agent swarms.
Why It Matters
Autonomous AI swarms in defense, cloud orchestration, and software engineering require flexible self-organization. If agents cannot autonomously allocate subtasks based on context, agentic systems cannot scale efficiently.
Research Objective
Analyze spontaneous specialization, leadership emergence, and consensus protocols in unconstrained multi-agent LLM systems placed inside shared workspace environments.
Technical Approach
Key Methodology Vectors
- Simulated multi-agent communication networks with shared memory channels and token budget constraints.
- Message trajectory analysis to measure role convergence, dominance metrics, and task completion speed.
- Ablation testing on communication protocols to identify parameters that stabilize or collapse consensus.
Expected Outcomes & Milestones
- Empirical protocols for self-organizing multi-agent swarms with minimal central orchestration overhead.
- Mathematical boundaries mapping when unconstrained multi-agent collaboration degrades into communication loops.
References & Prior Work
- Emergent Communication in Multi-Agent Reinforcement Learning — Journal of Artificial Intelligence Research (2024).
- Consensus Protocols for Autonomous Agent Swarms — Sleepers Research Lab Technical Memorandum (2026).
Interested in this research direction?
Sleepers Research welcomes inquiries from teams exploring autonomous agent security and applied machine learning.
Contact the Laboratory