Redirecting to original paper in 30 seconds...
Click below to go immediately or wait for automatic redirect
📄 Abstract
Abstract: Multi-agent LLM systems fail to realize parallel speedups due to costly
coordination. We present CodeCRDT, an observation-driven coordination pattern
where agents coordinate by monitoring a shared state with observable updates
and deterministic convergence, rather than explicit message passing. Using
Conflict-Free Replicated Data Types (CRDTs), CodeCRDT enables lock-free,
conflict-free concurrent code generation with strong eventual consistency.
Evaluation across 600 trials (6 tasks, 50 runs per mode) shows both benefits
and trade-offs: up to 21.1% speedup on some tasks, up to 39.4% slowdown on
others, and 100% convergence with zero merge failures. The study formalizes
observation-driven coordination for stochastic LLM agents, revealing semantic
conflict rates (5-10%) and quality-performance tradeoffs, and provides
empirical characterization of when parallel coordination succeeds versus fails
based on task structure.
Submitted
October 18, 2025
Key Contributions
CodeCRDT introduces an observation-driven coordination pattern for multi-agent LLM systems, replacing explicit message passing with monitoring of shared, observable states using CRDTs. This enables lock-free, conflict-free concurrent code generation with strong eventual consistency, addressing costly coordination issues and achieving speedups on some tasks while formalizing coordination strategies for stochastic agents.
Business Value
Enables more efficient and scalable development of complex software systems by leveraging multiple AI agents, potentially reducing development time and costs.