Coffeehouse Congregation Content System

A content operations system built to produce a steady volume of branded content without surrendering editorial judgement to the model. Batch generation does the volume; a review workflow keeps a human accountable for everything that ships.

AI content pipelines fail in one of two directions: either a human reviews every item and the throughput advantage disappears, or nobody reviews anything and quality collapses. A specific failure surfaced in practice — generated items that had errored out were still being swept into bulk approval, so broken output could reach production marked as reviewed.

Treat review state and generation state as separate concerns. Every generated item carries both its production status and its review status, and the bulk actions operate strictly on the intersection the reviewer intended. Failed items stay actionable — a reviewer can reject or regenerate them — but can never be silently approved in a batch.

A batch run pipeline with per-item approve, reject, and revise actions, a bulk approval path that excludes failed and already-reviewed items, and clear surfacing of generation errors alongside the actions available for them.

Human-in-the-loop workflow design, batch processing, AI content generation, review state modelling, operational UX.

Working system. Batch throughput holds up while every published item carries an explicit human decision behind it.

That the interesting design work in AI content systems is the review layer, not the generation — deciding exactly where human judgement is required and making it impossible to bypass by accident.