The author developed an autonomous AI-driven development factory designed to automate routine coding tasks. The system uses AI agents, referred to as gnomes, which operate independently and only escalate issues to a human when they encounter unsolvable problems.
The architecture relies on a core that manages the domain model and process flow, utilizing ports and adapters to integrate with existing tools like GitHub, Jira, and Jenkins. Developers can create custom adapters as plugins to support specific task trackers.
Process stages are defined using the IDEF0/ICOM model, with configurations stored in YAML files for stages and pipelines. This allows for flexible modifications to the development workflow by updating the configuration files without rewriting the core logic.
When an agent encounters an issue, it marks the task with a specific status in the tracker and adds a comment. Once the human provides feedback, the system resumes the task. An upcoming arbiter mechanism will use predefined criteria to make decisions, further reducing the need for human intervention.
The project requires Claude Code, OpenSpec, Java 25, and Git to function. The source code and a HelloWorld testing project are available for public use and modification.
