Show HN: Stigmergy pattern for multi-agent LLMs (80% fewer API calls) (github.com)

A groundbreaking approach has been introduced in multi-agent AI systems, utilizing the principle of stigmergy—similar to how ant colonies operate—to enhance collaboration among AI agents in software development. This innovative system features four independent AI agents: THINKER, BUILDER-UI, BUILDER-DDD, and GUARDIAN, which execute tasks without direct communication, coordinating solely through shared files in a Git repository. By leveraging Git’s conflict management, the project achieves a remarkable 80% reduction in token usage through incremental context loading, while maintaining uptime and offering self-improvement capabilities through pattern recognition. This method is significant for the AI/ML community as it simplifies multi-agent coordination, eliminating the need for complex messaging protocols and reducing overhead related to agent crashes or coordination rewrites when scaling. The agents autonomously claim, implement, and review tasks while utilizing a decentralized approach that not only enhances fault tolerance but also facilitates continuous learning. As AI deployments become more common, such efficient coordination could lead to lower operational costs and more robust systems capable of evolving autonomously, setting a new standard for AI-assisted software development.