Build an AI Agent Team That Ships Content

One agent asked to "write content" has to research, draft, edit, and publish all from a single set of instructions. Results are uneven. Splitting the work across a few agents, each with one job, produces better output and lets you check the work at each handoff.

This lesson builds that team in the Cosmic dashboard. No code, no configuration files. You will hire agents from a template library, run one, then connect several into a workflow that runs on a schedule.

---

What You Will Build

• An agent for content work, hired from a template and running in a few clicks

• A second agent that handles a different part of the job

• A workflow that runs them in order and hands work from one to the next

• A schedule so it all happens without anyone starting it

---

Before You Start

• A Cosmic account with a project and at least one bucket

• A content type in that bucket for the agents to write into, for example Blog Posts

• Project admin access. Agents and Workflows only appear in the sidebar for project admins

---

Step 1: Hire Your First Agent

You do not have to build an agent from nothing. Cosmic ships a library of ready-made agents with their instructions, model, and permissions already set.

1. In your project sidebar, click Agents

2. Click New Agent

3. Stay on the Use a template tab, which is where the page opens

4. In Search agents by name, role, or keyword, type

5. Click the card for Ava, Content Lead, described as Create high-quality blog posts, landing pages, and marketing content

6. In the panel that opens, click Hire Ava

Ava is now in your project with a working prompt and a sensible set of permissions, and hiring her drops you straight into a chat with her.

Browse the other templates while you are there. Elena, SEO Analyst audits content and optimizes metadata. Maya, Social Media Coordinator turns published posts into social copy. Lina, Translator & Localization Manager handles other languages. There are 28 templates across categories including Content & SEO, Growth & Analytics, and Operations.

If nothing fits, the Build from scratch tab gives you an empty form, and the Create with Agent button lets you describe the agent you want in plain language and have one built for you.

---

Step 2: Point the Agent at the Right Content

Everything about how an agent works lives in its settings, so open your new agent's Settings and check two things on the General tab.

Bucket decides which content the agent can see and write to. If your project has more than one bucket, make sure this is the right one.

Agent Prompt holds the agent's standing instructions. The template fills this in, and you should add the specifics only you know: your subject matter, your audience, anything the agent should never do. A few sentences is plenty.

That last line matters more than it looks. Without it, a vague request produces a confident draft about the wrong thing, which takes longer to catch than it saved.

---

Step 3: Check What the Agent Is Allowed to Do

Open the Tools & Capabilities tab and find the Capabilities section. These checkboxes are the agent's permissions, grouped into Content & Data, Agents & Automation, External Services, Analytics, and Repository.

The ones that matter for content work:

Templates come with a reasonable set already checked. The habit worth building is to uncheck anything the agent does not need rather than leaving everything on.

This is the real safety control. An agent without Write CMS content cannot change your content, no matter what its instructions say or what someone asks it to do in a chat. The prompt is guidance; the checkboxes are enforcement.

---

Step 4: Run It Once and Read the Result

Do not build the full pipeline yet. Give the agent one task and see what comes back.

Ask it for something small and specific, such as five topic ideas that your existing posts do not cover. Then read the output properly. You are checking whether the agent understood your subject matter, not whether the text is well-written. Well-written and wrong is the failure mode to watch for.

If the output misses, adjust the Agent Prompt and run it again. Two or three rounds here saves far more time than fixing drafts later.

---

Step 5: Add a Second Agent

Once the first agent is useful, hire a second one for a different part of the job. A common pairing is a researcher that finds and briefs topics, and a writer that turns a brief into a draft.

Keeping them separate is what makes the setup debuggable. When a draft comes out wrong, you can see whether the research or the writing caused it.

---

Step 6: Connect Them Into a Workflow

A workflow runs agents in order, one after another.

1. Click Workflows in the project sidebar

2. Click Create New Workflow

3. Fill in Workflow Name, for example

4. Optionally add a Description and a Workflow Prompt. The workflow prompt is included in every step, so it is the place for rules that apply to the whole run, like tone or subject matter

5. Under Workflow Steps, click Add Agent Step

6. Choose Use an existing agent and pick your researcher

7. In Instructions for this step (optional), describe what this step should do

8. Click Add Agent Step again and pick your writer, with its own instructions

9. Click Create Workflow

You Do Not Have to Wire Anything Together

This is the part people expect to be hard, and it isn't. Each step automatically receives what the previous steps produced. The writer step sees the researcher's output without you connecting them, naming a variable, or mapping any fields.

So step instructions can simply refer to the work already done:

Because agents are handing off in their own words, be explicit in each step about what the next one needs. "Output a brief with a working title, the reader question it answers, and three points to cover" gives the next agent something to work with. "Do research" does not.

Add a Human Checkpoint

Click Add Approval Gate to insert a step that pauses and waits for a person before continuing. Putting one before anything that publishes is a good habit while you are learning what the pipeline produces.

---

Step 7: Put It on a Schedule

In the Schedule section of the workflow, turn scheduling on. You will see a field labeled Describe your schedule.

Type when you want it to run in plain language:

Click Generate and Cosmic works out the timing for you. Set your Timezone and you are done.

If you would rather pick from options, click Switch to manual scheduling for a start time and a repeat dropdown with Every Hour, Every Day, Every Week, and Every Month.

A scheduled workflow can still be run by hand at any time from the Workflows page.

---

Reading What Happened

Every run is recorded. Open the workflow and you can see each step, whether it succeeded, how long it took, and what it produced.

When a run goes wrong, find the first step whose output looks off and fix that step's instructions. It is almost always one step, not the whole pipeline.

Drafts land in your CMS with status Draft. Nothing publishes without a person unless you deliberately set it up that way.

---

Do This in Your Project

Hire one agent and run it. That is the whole task.

1. Go to Agents and click New Agent

2. Search for and hire Ava, Content Lead

3. Confirm the Bucket is right, and add two or three sentences to the Agent Prompt about your audience

4. Ask it for five topic ideas your site does not cover

5. Read what it gives you

If the ideas are good, hire a second agent and put the two into a workflow. A two-step workflow you understand beats a five-step one you cannot follow.

---

Up Next

Lesson 2: Connect Cosmic to Cursor / Claude Code with MCP

Lesson 2 is for developers, and it goes the other direction: giving your AI coding tools direct access to the same content, so you can read and edit it from your editor.