Agency creatives open a full first draft of the brief, the calendar, the copy and the images in under ten minutes instead of a blank page.
The orchestrator delegates all four. The images stop and wait for a person.
The situation
How this used to work
Cambo runs campaigns for a roster of client brands. Every campaign passed through three or four people before anything went live: an account manager, a creative, a copywriter, a designer.
The brief was written from scratch. Then the content calendar. Then the copy for each post. The coordination between those people happened over email and scattered documents, which is where the time actually went.
What we built
The system
CambOS. Five agents, each with one job.
An account manager orchestrates. It delegates and it writes nothing itself, which is the point: an orchestrator that also drafts stops delegating and becomes the bottleneck it replaced. It does not generate images either. It hands that to the agent whose job it is.
Behind it, a brief creator, a calendar creator, a copywriter that writes Greek, and an image generator.
Underneath sits a Google Workspace server we built, so the output lands in the client's own Drive folder as a real document. Not in a chat window that somebody then has to copy out of.
How it works
The pipeline, step by step
One sentence, in their own words
A campaign for this client, promoting this product, the goal is awareness. That is the whole input.
The orchestrator hands out the work
It reads the request and calls the agents that are needed. It never drafts anything itself, and it never draws anything itself.
Brief, calendar, copy
The brief goes into a document. The calendar comes back with posting dates on it. The copy comes back per post, in Greek. The calendar creator hands the posts to the copywriter in small batches rather than asking for forty at once.
The image agent reads the calendar first
It opens the calendar sheet, finds the posts still missing an image, and works out the format each one needs from what the post actually is. A story is 9:16, a feed post is 4:5, a banner is 16:9. Then it goes looking in the client's own Drive for the real product mock-ups and brand assets to generate against, so the output belongs to the brand instead of resembling it.
The ratio comes from what the post is. The references come from their Drive.
It shows the plan and waits
Before a single image is generated, the agent presents what it intends to make and stops, whether that is one image or fifty. Nothing is drawn until a person says yes. The generated image goes back into the calendar row it belongs to, next to the post it illustrates.
Straight into their Drive
Every document writes into the client's own folder, in the tools the team already had open. Nobody exports anything.
Real documents in the client's own folder, not text in a chat window.
What changed
The result
The brief, the calendar, the copy and the images now arrive as drafts, in the client's own folder, before anyone has opened a blank document.
The team's job moved from making the first version to judging it. That is a smaller change than it sounds and a better one: the taste stays with the people who have it. The approval gate in front of the image generator is the same idea written into the software.
The stack
Agents
- Claude
- 5 agents
- hub-and-spoke + handoff
Images
- Nano Banana Pro
- Gemini 3.0 Pro Image
- Drive mock-ups as reference
- approval gate
Built by us
- Google Workspace MCP server
- Drive, Docs and Sheets tools
- Railway
Frontend
- React
- TypeScript
- Vite
- Supabase