What I learned building a native console for two competing coding agents

Developer Cyl Castillo shares insights from building 'Agent Console,' a desktop application designed to supervise coding agents like Claude Code and Codex. Moving beyond standard IDEs, the project focuses on providing a 'cockpit' for human oversight, featuring approval modals, diff viewers, and a verifiable audit trail for AI actions. Castillo highlights the architectural benefits of supporting multiple engines, noting that it forces developers to identify core abstractions rather than baking in agent-specific philosophies. The project also serves as a case study in the challenges of cross-platform development, including silent hook failures on Windows and clipboard inconsistencies in Linux. By utilizing a recursive development loop where the agents helped build their own management tool, the project emphasizes the importance of evidence-based trust in AI workflows. The tool is open-source, local-first, and aims to provide a stable interface for developers working daily with autonomous coding assistants.