The bottleneck in multi-agent coding is the human, not the agents (tskulbru.dev)

A recent exploration into multi-agent coding practices revealed that the true bottleneck isn't the agents themselves but rather the human operators managing multiple threads of work. The author, navigating complex coding tasks across various repositories, faced challenges in tracking progress and dependencies when using existing tools like GitHub. While Claude, an AI assistant, efficiently handled individual GitHub issues, it lacked awareness of interconnected tasks and context that resided outside of GitHub, leading to decreased autonomy for AI agents. To address this inefficiency, the author turned to Beads, a work graph built specifically for such coding agents. This tool acts as a local database that captures work items, their statuses (ready, blocked), and notes, thereby providing clearer visibility into ongoing tasks without cluttering public issue tracking platforms like GitHub. Beads integrates with external references, allowing team members to operate on a shared database while maintaining their individual notes and tracking subtasks without overwhelming one another. This approach not only streamlines task management but also enhances collaborative efforts in coding, marking a significant development in improving agent autonomy and task orchestration within AI-driven workflows.