
TableKeeper is a production-oriented restaurant reservation system designed to make table booking reliable even when real-world systems become complicated. Users can search for available restaurant tables for a specific time, create reservations, receive booking confirmations, and later modify or cancel those reservations. Behind the simple booking experience is a system designed around some of the hardest problems in reservation software: concurrent requests, duplicate retries, stale client state, time-zone handling, and maintaining consistent availability. A major focus of TableKeeper is ensuring that the same table cannot be accidentally double-booked when multiple users attempt to reserve it simultaneously. Operations are designed to be idempotent so retried requests do not create duplicate reservations, while related updates are handled atomically to keep system state consistent. The project evolves beyond basic CRUD operations by supporting reservation state over time. Changes to restaurant rules or reservation data can take effect at specific points without rewriting historical information, allowing past reservations to remain accurate and auditable. TableKeeper also supports larger operational changes where multiple existing records must be updated together without leaving the system in a partially updated state. What makes the project especially interesting is how it was built. Instead of using a traditional single-agent coding workflow, the application was developed through a BAND Dark Factory composed of multiple autonomous coding-agent seats. Agents were assigned different responsibilities for planning, implementation, review, verification, and hand-offs, creating a collaborative AI software-development workflow where generated work could be independently checked before acceptance. The result demonstrates both a robust reservation system and a practical experiment in autonomous multi-agent software engineering.
6 Oct 2026