Kiln — Autonomous Software Factory

Kiln is an autonomous software factory built with BAND to demonstrate a different approach to AI-assisted software development: instead of asking one AI agent to write an application, Kiln coordinates specialized agent seats that collectively take software from a product mandate to a verified build. The factory follows a structured workflow: Plan → Test → Build → Attack → Repair → Review → Integrate → Verify. For the Dark Factory challenge, Kiln was used to build Tablekeeper, an OpenTable-style restaurant reservation application. Tablekeeper allows users to discover restaurants, choose a date and party size, inspect availability, select a table, make a reservation, receive a confirmation reference, and retrieve their booking. The factory separates responsibilities across specialized seats including Planner, Test Author, Implementer, Investigator, Reviewer, and Integrator. This separation means the agent responsible for implementing functionality is not the only agent deciding whether that functionality is correct. BAND provides the collaborative agent-room environment through which these specialized roles operate, while Kiln provides the factory process, mandates, evidence trail, and validation workflow. The result is not only a working application, but an auditable development process. The repository contains the factory definition, agent mandates, BAND room export, staged implementations, acceptance work, and validation evidence. Tablekeeper reached Stage 4 with all four shipped-stage checks passing in isolated Docker validation. The final product is also deployed publicly for demonstration. This project explores what software development can look like when AI agents are organized as an engineering organization rather than used as a single coding assistant.