A nine-agent pipeline that turns a brief into a complete brand pack — with no human review. into a complete brand pack — with no human review.

Marketing Lab bridges the gap between $50/month Canva DIY and $50k agency engagements. Drop a client brief in 00_inbox/READY and a watcher fans out nine specialised Claude Code agents — identity, copy, print, website, social, video, plus a brief-synthesizer and a QA-guardian — against a single brand.json source of truth.

Each producer reads only brand.json + normalised assets, and writes only to its own folder. That isolation is what lets QA re-trigger one producer without re-running the whole pipeline. First test pack shipped 22 deliverables in 12 minutes, no human review per pack.

What’s actually hard. Not the agents — the quality fences. Industry-aware tonesets pick fonts, weights and composition from the brand.json industry field (athletic / artisan / professional / hospitality). Real-logo preservation refuses to redraw a client’s mark, only normalises it. Multi-line text-fitting measures Slovak headlines character-by-character to prevent clipping. Legibility guards add a gradient overlay or a white-with-shadow fallback when text sits over a photo region. Theme-based photo picker rotates fairly without reuse.

The QA-guardian reads the output folder, judges whether each deliverable passes the bar, and re-triggers only the producer that failed — not the whole pipeline. That’s what makes the system economically real.