Gather — Built by a Dark Factory

Gather is a restaurant reservation platform created by an autonomous software factory for the Tablekeeper track of the WeAreDevelopers × BAND Dark Factory Hackathon. Instead of treating AI as a coding assistant, we designed a reusable three-seat factory: a Coordinator plans and delegates work, an Implementer builds the solution, and an independent Reviewer challenges the implementation before accepting it. The mandates are generic and reusable rather than Tablekeeper-specific. The factory progressed cumulatively through all four Tablekeeper stages. Review was not ceremonial: implementations were rejected when defects were discovered, corrected through additive commits, independently re-tested, and only then accepted. In the final stage, the Reviewer identified consistency problems beyond the public test suite and required additional correction cycles before final acceptance. The resulting application, Gather, provides a complete restaurant reservation experience with availability discovery, table-capacity management, combined seating, authenticated reservations, confirmation references, reservation lookup and cancellation. Underneath the interface, the system handles harder concerns including concurrency, idempotency, time zones, recurring reservations, policy evolution and atomic seating replanning. The final cumulative Stage 4 verification passed all official stage checks. Additional independent adversarial testing exercised optimizer behavior, rollback guarantees, stale-plan handling, concurrency, API behavior and browser flows. Most importantly, Gather is not simply a demonstration of what an AI model can code. It demonstrates a software factory in which autonomous agents coordinate, implement, review, reject, correct and ultimately ship working software.