AI for Marketing: Build a Multi-Agent Content Pipeline With Slack

The usual way to use AI for content writing is to paste a topic into a chatbot and get back 1,200 words of confident filler. The problem isn't the model. It's that one prompt is being asked to do three jobs, research, writing and editing, that a content team keeps separate for good reasons. This guide splits them across three agents in a Damon workflow. They pick up the ideas your team drops in Slack and hand you a researched, on-brand draft with an editor's notes every Monday.

Nothing gets published by the pipeline. It ends with a draft in your drive and a message to you.

What you get

Every Monday at 9:00, a thread like this:

This week's draft: "Why annual plans beat discounts for SaaS retention", from Priya's idea in #content-ideas on Thursday. 1,340 words, in Content/2026-10-05-annual-plans/draft.md. Editor's notes: two claims tightened to match the sources, intro cut from three paragraphs to one, one statistic flagged as needing a primary source before publishing. Suggested LinkedIn post and meta description are at the bottom.

Next to the draft in the drive sit the researcher's brief, with every source it used, and the editor's notes. You can see how the draft was made, not just what came out.

Why three agents instead of one

• The researcher reads the ideas channel, picks a topic, searches the web and writes a brief with sources. It writes no prose, so it doesn't pad.
• The writer only sees the brief and your brand voice. It can't cite something the researcher didn't find, which cuts down on invented statistics.
• The editor checks the draft against the brief and your style rules, fixes what it can and flags what it can't. A fresh reader catches what the author can't.

Each is a separate Damon agent with its own personality and permissions. The workflow runs them in sequence, and each step receives the previous step's output. (New to agents? Start with how to create an AI agent.)

Step 1: Connect Slack and write down your voice

In Integrations, connect Slack. The pipeline only reads it: the channel history of #content-ideas, or wherever your team drops ideas. Reads are auto-approved by default. Posting to Slack is message.send, which requires approval, and the core pipeline doesn't post.

Then put your brand voice in the drive, where every agent can read it: Content/brand-voice.md. Include your audience, tone, words you never use, heading style, target length and two paragraphs you consider on-brand. A page is enough. This file does more for quality than any prompt.

Step 2: Create the three agents

In Agents (or ask Damon to create them):

• Researcher, @research
• Personality: "Find primary sources. Summarize, don't write prose. Every claim in the brief has a link. Say when evidence is thin."
• Integrations: Slack.
• Writer, @writer
• Personality: "Write from the brief only. Follow Content/brand-voice.md exactly. Concrete examples over general statements. No claim without a source in the brief."
• Integrations: none needed; it works in the drive.
• Editor, @editor
• Personality: "Cut first. Check every factual claim against the brief. Fix style issues directly; flag anything you can't verify rather than softening it."
• Integrations: none needed.

Step 3: Run it once in chat

Try each role by hand before automating, so you can tune it:

@research Read #content-ideas from the last 7 days. Pick the idea with the most engagement that fits Content/brand-voice.md's audience. Research it on the web and write Content/<date>-<slug>/brief.md: the angle, 5–8 key points each with a source link, and what competitors' articles on this topic miss.

Then ask @writer to draft from that brief, and @editor to edit the draft. Adjust each agent's personality until you'd be happy with its output alone. Most fixes go into brand-voice.md, not the agents.

Step 4: Make it a weekly pipeline

The workflow:

1. Trigger: every Monday at 9:00.
2. Agent step (@research): reads #content-ideas for the last 7 days, skips ideas already covered in Content/, picks one, researches it and writes brief.md.
3. Agent step (@writer): writes draft.md from the brief and the brand voice.
4. Agent step (@editor): edits the draft in place, writes editor-notes.md with what it changed and what it couldn't verify, and appends a suggested meta description and LinkedIn post.
5. Message step: opens a thread with the title, the source idea, the file paths and the editor's notes.

After cloning the template, open each agent step and choose its agent. Templates ship with the default agent in every step, and the split into three only pays off once each step has its own personality.

The agents write to your drive, which is file.write and requires approval by default. Auto-approve file.write for this workflow so the pipeline runs unattended. That permission applies to this workflow only, not to the agents everywhere.

Variants

• Team review in Slack. Add a final step that posts the title and draft link to #content-review. That's a message.send. Auto-approving it for that one step is reasonable when its only instruction is to post to an internal channel, since that's your team, not your audience.
• A repurposing agent. Add a fourth agent that cuts the edited draft into a LinkedIn post, an X thread and a newsletter blurb, in social.md next to the draft.
• Ideas from customers. Point the researcher at support or sales call transcripts (Fireflies) instead of a Slack channel, so topics come from questions customers actually ask. The feedback synthesis recipe uses the same sources.
• Two drafts a week. Change the trigger to twice weekly, or have the researcher pick two ideas and let each step handle both.

Common questions

Will it publish to our blog or social accounts?

No. The pipeline writes drafts to your drive and messages you. Publishing is a person's decision; posting to external channels would be a message.send, which requires approval, and we recommend keeping it that way for anything your audience sees.

How do I stop it making things up?

The pipeline's structure is the main defense. The writer only works from the researcher's brief, every point in the brief carries a link, and the editor checks the draft against the brief and flags what it can't verify. Read the editor's notes before publishing; they're short.

Can the agents use our existing style guide?

Yes. Put it in the drive (a PDF or Word file works; Damon converts it) and reference it from brand-voice.md. Or save the rules as a skill so they also apply when you ask for copy in chat.

Does it need three separate agents?

It runs with the default agent in every step, but separate agents are the point: each has its own personality and instructions, and the editor genuinely reads the draft as someone else's work.