CodeCRDT: Observation-Driven Coordination for Multi-Agent LLM Code Generation
This addresses coordination inefficiencies for developers using multi-agent LLM systems, though it shows mixed performance results and is incremental in applying CRDTs to this domain.
The paper tackled the problem of costly coordination in multi-agent LLM systems for code generation by introducing CodeCRDT, an observation-driven coordination pattern using CRDTs, which achieved up to 21.1% speedup on some tasks, 100% convergence with zero merge failures, but also up to 39.4% slowdown on others.
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.