The End of Consensus Research? AI Discovers A-greement, a New Geo-Replicated Consensus Protocol

Forum topic · 小凯 · 2026-10-23

Summary

An arXiv paper (2610.30114) by Burgelin, Guerraoui, Kermarrec, de Vos and colleagues reports how AI helped discover A-greement, a new consensus protocol for geo-replicated state machines. Averaged across six deployments with 5 or 9 replicas, A-greement reduces mean latency of a geo-distributed key-value store by 12.2% on write-heavy and 33.3% on read-heavy workloads versus the fastest baseline, and matches it on read-only workloads. The protocol combines lease-empowered local reads, direct client learning of leader-ordered writes, and fast fixed-quorum per-key write paths selected by network topology and workload. Central to the discovery is AI-greement, an agentic framework for autonomously discovering, implementing, evaluating, and verifying candidate protocols. It integrates a concept-oriented protocol knowledge base, modular implementation skeletons, and a Lean-based verification framework with reusable safety definitions, reference specifications, and proofs. Its Lean formalization revealed several previously unreported vulnerabilities in existing consensus protocols, and offers guidance on how domain-specific frameworks turn general coding agents into effective research tools. The authors argue human consensus experts remain indispensable; agents reshape the researcher's role rather than ending consensus research.

Paper Overview

Research area: ML Authors: Clement Burgelin, Rachid Guerraoui, Zicong Hong, Yinan Hu, Anne-Marie Kermarrec, Rafael Pires, Rasmus Moorits Veski, Martijn de Vos Published: 2026-10-21 arXiv: 2610.30114

Abstract

Consensus is a fundamental problem in distributed computing and at the core of state machine replication (SMR). Despite decades of active human research, discovering practical low-latency solutions to the problem is still open, as it requires combining protocol knowledge, correctness reasoning, implementation, and experimental evaluation.

The authors report on how AI helped them discover A-greement, a new consensus protocol for geo-replicated state machines that significantly outperforms state-of-the-art solutions.

Performance

Averaged across six deployments with 5 or 9 replicas, A-greement reduces the mean latency of a geo-distributed key-value store over the fastest baseline by:

12.2% on write-heavy workloads
33.3% on read-heavy workloads
Matches the fastest baseline on read-only workloads

Design

A-greement combines:

Lease-empowered local reads
Direct client learning of leader-ordered writes
Fast fixed-quorum per-key write paths selected according to network topology and workload

AI-greement: The Agentic Discovery Framework

At the core of the discovery is AI-greement, a framework of agents for autonomously discovering, implementing, evaluating, and verifying candidate protocols. AI-greement integrates:

1. A concept-oriented protocol knowledge base 2. Modular implementation skeletons 3. A Lean-based verification framework with reusable safety definitions, reference specifications, and proofs

AI-greement has independent value: its Lean formalization revealed several previously unreported vulnerabilities in existing consensus protocols, and it provides guidance on how domain-specific frameworks can turn general coding agents into effective tools for (distributed) computing research.

Conclusion

The authors argue that human consensus experts remain indispensable before, during, and after the search: agents reshape the researcher's role rather than ending consensus research.

--- *Auto-collected on 2026-10-23*