AI multi-agent systems advance automated code review capabilities

Developments in AI-driven software development are shifting toward multi-agent systems for automated code reviews. A comparison between Claude Code’s ultrareview and Dromeas’s LLM council highlighted different approaches to identifying code issues. In a test involving a large, human-approved pull request from the OpenClaw repository, Dromeas utilized three independent models—Opus 5, DeepSeek V4 Pro, and GPT-5.6 Terra—to analyze quality, security, and compliance, resulting in 17 verified findings.

Parallel to these commercial and specialized tools, independent developers are experimenting with agentic architectures. One project, the PR Review Crew, utilizes a six-agent system orchestrated via ChatDev 2.0. This system employs specialized agents for fetching diffs, reviewing logic, security, and documentation, and synthesizing findings into a single GitHub comment. These projects emphasize the importance of explicit context propagation and the separation of deterministic actions from reasoning in multi-agent workflows.