CEPO Splits Coordination from Execution, Boosting Multi-Agent Prompt Optimization Across 11 Benchmarks
Jongha Kim · hf · 2026-11-02
A paper from Korea University, KAIST and Google Cloud AI Research introduces CEPO (Coordinator-Executor Prompt Optimization), a general framework for prompt optimization in multi-agent systems.
- Problem: Existing prompt optimizers couple system-level coordination with agent-level execution inside each agent's prompt, so coordination rules can't be refined consistently across agents.
- Method: CEPO splits prompts into a shared coordinator prompt plus per-agent executor prompts, updating each according to its role.
- Results: Across 11 benchmarks in four domains, applying CEPO to four existing prompt optimizers consistently improves performance, with better optimization-budget efficiency and scaling as agent count grows.
More from coding & agent
- Open-Source TypeScript MCP Server Template With Fastify, Docker and Full Dev Tooling — modelcontextprotocol · 2026-11-02
- BLS MCP Server Connects AI Models to Bureau of Labor Statistics Public Data API v2 — modelcontextprotocol · 2026-11-02
- Dev builds 2-agent autonomous GTA 6 recon system on Cloudflare with free models — MicahBerkley · 2026-11-02
- RUC open-sources FlashRSI: one framework unifying 13 agent self-improvement algorithms — aigclink · 2026-11-02
- Claude Code and Antigravity Both Emergently Learned to Run Risky Experiments as AX Tasks — rakyll · 2026-11-02
- Evals want step-function tasks, training wants smooth pass@k curves — why the two goals conflict — JoshPurtell · 2026-11-02