Cosmic AI
April 21, 2026
Every content team has the same problem: not enough time, too many channels, and a publishing backlog that never clears.
AI writing tools help with drafts. But they don't research. They don't publish. They don't generate images, schedule posts, or respond to Slack requests. You still do all of that yourself.
A Cosmic AI agent changes the equation. It is not a writing assistant. It is an autonomous agent that can research topics on the web, write full-length posts, generate cover images, publish directly to your CMS, and run on a schedule without anyone pressing a button.
This post covers exactly how that works: what a Cosmic Content Agent is, what it can do, how to configure one, and how to chain it into a multi-step workflow that covers your entire content pipeline from idea to published post.
What a Cosmic AI Agent Actually Is
Before getting into configuration, it helps to understand what kind of software this is.
Cosmic ships four distinct agent types:
Content Agents generate and manage CMS content. They can read your existing objects, write new ones, update metadata, generate AI images, generate AI video, and publish or schedule content. They run in response to manual triggers, webhooks, or on a cron schedule.
Team Agents live in Slack, WhatsApp, and Telegram. They have persistent memory across conversations and can take on custom personas and goals. A Team Agent in your Slack channel can receive a request like "write a blog post about headless CMS for Astro" and execute the full pipeline: research, write, image, publish.
Code Agents connect to GitHub repositories, create branches, commit code, and open pull requests. A content pipeline can hand off to a Code Agent to build a new landing page after a post is published.
Computer Use Agents control browsers using visual AI. They can navigate websites, extract data, cross-post content, and record demo videos.
For content pipeline automation, you will primarily use Content Agents (scheduled, autonomous publishing) and Team Agents (conversational, on-demand). This post focuses on both.
What Can a Content Agent Do in a Pipeline?
Here is the realistic scope of what a Cosmic Content Agent can handle autonomously:
Research: With web browsing capability enabled, an agent can fetch URLs, read competitor posts, analyze search trends, and pull in real facts before writing. It does not hallucinate sources because it fetches them.
Write: The agent writes full-length blog posts, comparison pages, tutorials, product announcements, and social copy. You define the format, tone, and structure in the system prompt.
Generate images: The agent can call image generation to create hero images, inline illustrations, and thumbnails. It uploads them to your media library automatically.
Publish to CMS: The agent creates Objects in your Cosmic bucket with all the right metadata: title, slug, teaser, author, category, tags, published date, and status. It can publish immediately or set to draft for review.
Send notifications: After publishing, the agent can ping a Slack channel, send an email, or message a Telegram chat with the post URL.
Run on schedule: Agents can run hourly, daily, weekly, or on any cron expression. Free plan agents run manually only. Paid plans (Builder and above) get full scheduling.
Setting Up Your First Content Agent
Step 1: Create the Agent
In your Cosmic dashboard, navigate to AI > Agents and click Create Agent.
Choose Content Agent as the type.
Give it a name like and write a system prompt. The system prompt is the agent's standing instructions. Here is an example for a blog publishing agent:
Enable the capabilities your agent needs:
- CMS Read/Write: required for creating and updating objects
- Web Browse: required for research
- Notify Send: required for Slack/email notifications
- Generate Images: for hero image creation
Step 2: Set a Schedule (Paid Plans)
Under the Schedule tab, enable scheduling and set a cron expression:
For the scheduled prompt, write what you want the agent to do each time it runs:
The agent runs autonomously at the scheduled time. No one needs to be logged in. No one needs to trigger it.
Step 3: Test It Manually First
Before enabling the schedule, run the agent manually once to verify it:
- Creates the object correctly with all required metadata
- Generates an image and attaches it
- Posts to your notification channel
- Produces content at the quality level you expect
Run it from the dashboard, watch the execution log, and refine your prompt based on the output.
Setting Up a Team Agent for On-Demand Publishing
A scheduled Content Agent handles your recurring publishing cadence. A Team Agent handles on-demand requests from your team in Slack.
Create the Team Agent
In AI > Agents, create a new Team Agent.
Connect it to your Slack workspace. The agent gets a Slack identity: a name, an avatar, and a channel to live in.
Example system prompt for a Slack-based content agent:
With persistent memory enabled, the agent remembers what it has published across conversations. Ask it "what have you written this month?" and it will tell you.
Using the Team Agent in Practice
Once connected, team members interact with it directly in Slack:
The agent researches, writes, generates the image, publishes the draft, and replies with the URL. The whole process takes a few minutes. The team member reviews and publishes from the CMS dashboard.
This turns your entire team into content contributors without requiring any of them to write.
Building a Multi-Step Content Workflow
Individual agents are powerful. Multi-step Workflows are where the full automation potential appears.
A Workflow chains multiple agents together, passing context from one step to the next. Each step can be a different agent type.
Example: Full Content Launch Pipeline
Here is a workflow that takes a product announcement from idea to published post to social distribution:
Step 1: Content Agent (Write and Publish)
Prompt: "Research [topic], write a 2,000-word blog post, generate a hero image, and publish as a draft. Return the post title, URL, and a 280-character Twitter summary."
Step 2: Content Agent (Social Posts)
Receives the post title and URL from Step 1.
Prompt: "Using the post title and URL provided, write social posts for Twitter, LinkedIn, and a Slack announcement. Include UTM parameters on all links."
Step 3: Team Agent (Notify)
Receives the social posts from Step 2.
Prompt: "Post the Slack announcement to the #content-team channel. Include the CMS draft link for review before publishing."
This three-step workflow takes a topic and outputs a drafted blog post, ready-to-use social copy, and a Slack notification. End to end. Fully automated.
Example: Competitor Intelligence Workflow
Step 1: Computer Use Agent
Navigates to competitor blogs, reads recent posts, and extracts titles, topics, and publish dates.
Step 2: Content Agent
Receives the competitor content list. Identifies gaps. Writes a brief report. Creates a prioritized list of topics to counter with Cosmic content.
Step 3: Team Agent
Posts the competitive intelligence report to #content-strategy in Slack.
This workflow runs every Monday morning. Your team starts the week with a ready-made competitive brief.
Webhook-Triggered Automation
Agents can also respond to webhooks, which means external events can trigger your content pipeline.
Examples:
- A new product feature ships. Your CI/CD pipeline sends a webhook to Cosmic. A Content Agent picks it up and drafts the feature announcement post.
- A customer submits a support ticket that your team escalates to a known bug. A webhook triggers a Content Agent to draft a knowledge base article.
- Your analytics system detects a post driving unusual traffic. A webhook triggers an agent to write a follow-up post on the same topic.
To set up webhook triggers:
- In your agent settings, enable Webhook Trigger
- Copy your agent's webhook URL
- Configure your external system to POST to that URL when the event fires
- The request body becomes the agent's input context
Webhooks require the Webhooks add-on ($99/month, or $199/month as part of the feature bundle).
Practical Prompt Engineering for Content Agents
The quality of your agent's output is directly proportional to the quality of your system prompt. A few principles that work:
Be specific about format. Don't say "write a blog post." Say "write a blog post between 1,500 and 2,000 words with an H2 section for each major concept, a summary table, and three FAQ questions at the end."
Define tone explicitly. "Clear, direct, developer-friendly. No buzzwords. No em dashes. Technical depth when relevant, but accessible to non-technical readers."
Specify what to avoid. "Do not fabricate statistics, case studies, or customer stories. Only cite sources you have verified by reading them via web browsing."
Include required CTA structure. "Every post must end with a CTA section linking to https://app.cosmicjs.com/signup and https://calendly.com/tonyspiro/cosmic-intro."
Define the metadata schema. Tell the agent exactly what fields to populate in the CMS: title, slug, teaser, author, category, tags, published_date, status. If you leave this vague, the agent will guess and sometimes get it wrong.
Content Quality Controls
Automation at scale needs quality checkpoints. A few approaches:
Draft-first publishing. Set your agent to always publish as , not . Your team reviews and publishes manually. The agent handles the writing; humans handle the editorial gate.
Review notification. After drafting, have the agent notify the content lead in Slack with a summary: post title, word count, key claims, and the CMS link. The reviewer can spot-check in under two minutes.
Approval gates in Workflows. Cosmic Workflows support approval steps. You can pause a workflow after Step 1 (drafting) and require a human to approve before Step 2 (social posting) runs.
Prompt versioning. Treat your system prompt like code. Keep a version history. When you change the prompt, note what changed and why. When output quality drops, you can identify the cause.
What This Looks Like in Practice
Here is a realistic weekly content operation for a team using Cosmic AI agents:
Monday 8am (automated): Competitor analysis workflow runs. Content strategy brief posted to Slack.
Monday-Friday (on-demand via Slack): Team members request posts from the Slack-connected Team Agent. Each request produces a draft in the CMS within minutes.
Wednesday and Friday (scheduled): Content Agent runs on schedule, picks a topic from a curated list stored in the CMS, writes and images the post, publishes as draft.
End of day (manual review): Content lead reviews all drafts queued in the CMS. Publishes or revises as needed.
After each publish (automated): A webhook triggers the social posting workflow. Social copy is posted to the team's review queue.
This pattern lets a one- or two-person content team operate at the output level of a much larger team. The agents handle the production work. The humans handle strategy and quality.
Frequently Asked Questions
Can a Cosmic AI agent publish content without human review?
Yes. You can configure the agent to set status to instead of . However, publishing as draft and reviewing manually is strongly recommended for any externally visible content.
How do I control what topics the agent writes about?
Two approaches: (1) Keep a "content briefs" Object type in your CMS with draft topic ideas. Configure the agent to pull from there. (2) Use the scheduled prompt to specify topic categories or source URLs for the agent to reference.
Do AI agents require a paid plan?
Free plan includes 3 agents that run manually. Scheduling requires a paid plan (Builder, $49/month, includes 5 agents with scheduling). Webhooks require the Webhooks add-on.
Can agents write in different tones for different content types?
Yes. Create separate agents with different system prompts: one for developer tutorials (technical, direct), one for thought leadership (opinionated, strategic), one for product announcements (concise, feature-focused).
How does the agent know what's already been published?
With CMS Read capability enabled, the agent can query your existing objects before drafting. Instruct it to check for existing posts on similar topics and avoid duplication. Team Agents with persistent memory also remember their own past outputs.
What AI models do the agents use?
Cosmic agents run on leading models including Claude Sonnet and Claude Opus. Model choice affects token consumption (Standard models use 2x tokens, Premium models use 4x, Reasoning models use 8x against your plan's token allowance).
Getting Started
The fastest way to get a content agent running:
- Sign up for Cosmic free (no credit card)
- Create a bucket and set up your blog post Object type
- Go to AI > Agents > Create Agent
- Choose Content Agent, write your system prompt, enable CMS Write and Web Browse
- Run it manually on a test topic and review the output
- Upgrade to Builder ($49/month) to enable scheduling
- Set a cron schedule and let it run
From zero to a publishing AI agent in under an hour.
For teams that want a more sophisticated setup with Slack integration, multi-step workflows, and webhook triggers, book a demo with Tony and we'll walk through it live.
The Bottom Line
Content pipeline automation is not about replacing writers. It is about eliminating the operational overhead that slows writers down: the research queues, the image requests, the scheduling friction, the republishing lag.
A Cosmic AI agent handles the production layer. Your team handles the strategy, the judgment calls, and the editorial quality. The result is a content operation that scales with your ambitions instead of your headcount.
Start free. Ship faster.
Continue Learning
Ready to get started?
Build your next project with Cosmic and start creating content faster.
No credit card required • 75,000+ developers
