---
title: "Agency OS: how a creative agency runs from brief to delivery with 27 AI agents"
description: "Agency OS, a creative agency run with AI agents: 27 agents across 8 teams, 9 workflows and a director sign-off at every key step, from brief to delivery."
url: https://addabra.com/en/blog/agency-os-creative-agency
lang: en
date: 2026-10-13
category: "Industries"
keyword: "creative agency with AI agents"
publisher: Addabra
alternate: https://addabra.com/blog/agency-os-agencia-creativa
---

# Agency OS: how a creative agency runs from brief to delivery with 27 AI agents

> Agency OS, a creative agency run with AI agents: 27 agents across 8 teams, 9 workflows and a director sign-off at every key step, from brief to delivery.

Thursday, 7 p.m. The account director at a creative agency receives a brief: a consumer brand wants an integrated campaign for a launch in six weeks. Before anyone has a single idea, someone has to reread a year's worth of emails with the client, dig up what was pitched and what got rejected, review the brand guidelines, request market data and coordinate strategy, creative and production. That usually takes days, and it depends on the person who "knows the client" being available.

Agency OS was built to solve exactly that. It's a system Addabra builds to run a **creative agency with AI agents**: 27 agents organized into 8 teams, 9 end-to-end workflows and a human director's sign-off at every key step. In this article we walk through how it's put together and what you can learn from it even if your company isn't an agency.

## The problem: creative talent stuck in coordination

At an agency, the value is in judgment: understanding the client, finding the idea, defending it and executing it well. Yet a big share of the best people's day goes to tasks that don't need that judgment.

- Rebuilding each client's context from old emails and decks.
- Translating the brief between teams: what accounts understands isn't always what creative needs.
- Producing variations and adaptations of assets for every format.
- Chasing approvals and versions.
- Measuring results and building the client report.

As the agency grows, that coordination grows faster than revenue. The traditional fix is hiring more account and production staff. Agency OS takes a different route: an agentic system handles the coordination and groundwork, and people sign off on what matters.

## A creative agency with AI agents: 27 agents across 8 teams

The system mirrors the structure of a real agency. Each team has agents with a specific, well-scoped job:

| Team | What its agents do |
|---|---|
| Accounts | Take in the brief, structure it and keep the client relationship up to date |
| Insight and strategy | Research market, audience and competitors, and propose the strategic approach |
| Brand | Make sure everything respects each client's identity, tone and guidelines |
| Creative | Generate concepts, territories and first versions of ideas |
| Production | Develop assets, variations and format adaptations |
| Measurement | Define KPIs and read campaign results |
| Operations and delivery | Coordinate timelines, versions and packaging of deliverables |
| Quality and memory | Audit the other agents' work and keep the agency brain up to date |

No agent does "everything." That's a design decision: an agent with a clear job is easier to evaluate, correct and improve.

## Nine workflows, from brief to delivery

The agents don't work in isolation. They're organized into 9 workflows, each with defined steps, dependencies and approval points. Two examples:

- **Integrated campaign.** From brief to full plan: insight, strategy, creative concept, assets by channel, measurement plan and client presentation.
- **Brand platform.** For clients who need to define or refresh their brand: diagnosis, purpose, positioning, personality, tone and guidelines.

Each workflow behaves like a graph: a step doesn't start until the previous one is approved, and if something fails, only that step is retried instead of redoing everything. We explain this logic in [multi-agent orchestration](https://addabra.com/en/blog/multi-agent-orchestration).

## The pieces that make it reliable

What separates Agency OS from "using AI at the agency" isn't the models. It's the harness: the structure around the agents that defines what they can do, with what information and under what controls.

### The agency brain: one file per client

All memory lives in one place. Each client has a file with brand, tone, audience, past campaigns, and what they approved, what they rejected and why. When a brief comes in, every agent starts with that context. When the client approves or rejects something, the file is updated. We cover this in detail in [one file per client](https://addabra.com/en/blog/one-file-per-client-memory).

### Human gates signed by role

At every key step there's a gate that only a person with the right role can open. The strategy director signs off on the approach; the creative director, on the concept; the account director, on the final presentation. Without that signature, the workflow doesn't move forward, and every approval is logged.

### A cross-auditor

The agent that produces an asset is never the one that reviews it. Agency OS uses an auditor model different from the one that did the work to check consistency with the brief, the brand and the strategy. It's the principle we describe in [generator and evaluator](https://addabra.com/en/blog/generator-and-evaluator-agents): nobody should grade their own exam.

### A spending cap per run

Every workflow run has a cost limit. If a campaign needs more iterations than planned and approaches the cap, the system stops and asks for authorization. That keeps operating costs predictable.

## What the client receives

At the end of a workflow like the integrated campaign, the system delivers concrete packages, ready for human review:

- **A PDF proposal** with insight, strategy, concept and measurement plan.
- **Assets** developed and adapted to the agreed formats.
- **A presentation deck** for the client meeting.

Everything has passed through each director's gate before it goes out.

## Illustrative example: one brief, before and after

Picture Thursday's brief at an agency without the system. Friday goes to gathering context. Monday there's an internal alignment meeting. Strategy delivers on Wednesday, creative starts on Thursday and the first internal presentation lands a week and a half later.

Now with Agency OS. Thursday night, accounts structures the brief and the system pulls together the context from the client file. First thing in the morning, strategy receives research and a preliminary approach to review and sign. With that signature, creative receives territories to choose from and develop. People spend their time deciding, adjusting and improving, not hunting for information.

## Mistakes we avoided while building it

- **Overly general agents.** A "creative agent" that does everything is impossible to evaluate.
- **Scattered memory.** If client context lives in random folders, every agent works from a different version.
- **Informal approvals.** A "looks good" in a chat isn't a gate. The sign-off must be explicit and tied to a role.
- **Self-evaluation.** Asking the same model to review its own work creates false confidence.

## What to do on Monday

If you run an agency, take your last project and log how many hours went to rebuilding context, coordinating and adapting assets, versus hours of strategic and creative thinking. That ratio will tell you how much room you have to grow without hiring at the same pace.

At Addabra we build systems like Agency OS tailored to each company, on our own proven harness: diagnosis, agent and workflow design, build, production rollout and ongoing evolution. The agency keeps its judgment and its signature; the system does the rest.

## Frequently asked questions

### Does Agency OS replace the agency’s creatives?

No. The agents research, organize, draft first versions, produce variations and run checks. Creative and strategic decisions are made by people, and every key step requires sign-off from the right role.

### What is the agency brain?

It is the system’s shared memory: one file per client with their brand, tone, campaign history, and what they approved and rejected. Every agent reads from it, so work starts with full context.

### How do you control the cost of running so many agents?

Every workflow run has a defined spending cap. If a campaign gets close to that limit, the system stops and asks for authorization instead of continuing to spend.

## About Addabra

Addabra is an AI-software-first software factory that designs and builds custom agentic systems for companies: one agent per process, a shared business memory, orchestration, human approval by role, auditing and cost control. Free diagnosis: https://addabra.com/en/blog/agency-os-creative-agency#contacto