My Newsletter Swarm: How 3 AI Agents Help Me Publish Every Week

By Caiado · 8 min read

I used to spend my entire Sunday writing my newsletter. Six hours of research, drafting, editing, fact-checking, and formatting. Every. Single. Week.

Today, I spend 90 minutes on Sunday morning. The rest happens automatically, coordinated by three AI agents that work while I sleep, work out, and spend time with my family.

This isn't theoretical. This is my actual workflow for the newsletter you're reading right now.

                TL;DR: I went from 6+ hours per newsletter to 90 minutes. The agents (Scout, Quill, Peter) handle research, drafting, and review. I focus on strategy, voice, and final approval.
            

The Before: My Manual Workflow

Here's what my Sunday used to look like:

9:00 AM — 12:00 PM: Research. Scrolling through 50+ sources, taking notes, saving links.
12:00 PM — 2:00 PM: Writing first draft. Usually interrupted by lunch.
2:00 PM — 3:00 PM: Fact-checking. Verifying quotes, double-checking statistics.
3:00 PM — 4:00 PM: Editing and formatting for Substack.
4:00 PM — 5:00 PM:> Creating social media threads, scheduling posts.

Total time: 6-7 hours. And honestly? The quality was inconsistent. Some weeks I was sharp; other weeks I was rushing to finish before dinner.

The After: My Agent-Assisted Workflow

Now the work is distributed across three days, with agents handling the heavy lifting:

🔍 Friday — Research Phase

🔍

Scout (Research Agent)

Scans 50+ AI news sources, tech blogs, research papers, and Twitter/X threads. Flags 10-15 stories with high signal-to-noise ratio. Saves summaries to shared workspace.

👤

Me (The Editor)

Review Scout's flagged stories over morning coffee. Pick 4-5 to cover. Send brief to Scout: "Deep dive these, extract key quotes, identify counter-arguments."

✍️ Saturday — Writing Phase

🔍

Scout (Deep Research)

For each selected story: reads original sources, extracts key quotes, identifies expert opinions, finds data/statistics. Compiles research briefs with citations.

✍️

Quill (Writer Agent)

Takes Scout's research and drafts initial sections. Structures arguments. Writes intro hooks. Generates headline options. Leaves [COMMENT] tags where my voice/input is needed.

👤

Me (The Voice)

Evening editing session. Read Quill's drafts. Add my commentary, opinions, and personal anecdotes. Rewrite sections to match my voice. Fill in [COMMENT] tags with hot takes.

✅ Sunday — Review & Publish Phase

📋

Peter (Review Agent)

Fact-checks every claim. Verifies statistics against original sources. Flags unsupported assertions. Suggests structural improvements. Catches grammar and style issues.

📢

Echo (Distribution Agent)

Formats for Substack. Creates social media thread drafts. Generates email subject line variations. Prepares LinkedIn post version.

👤

Me (Final Approval)

Morning review (90 minutes): Read Peter's flagged items. Address any fact-check issues. Final polish. Review Echo's social drafts. Hit publish. Done by 10:30 AM.

The Results

6-7 hrs Before (Manual)

90 min After (Agent-Assisted)

But it's not just about time. It's about quality:

More sources: Scout finds perspectives I'd miss on my own

Better fact-checking: Peter caught 3 errors last month that I would have published

Consistent voice: Quill maintains structure so my edits are about flavor, not fixing messes

Less burnout: I'm not fried after publishing. I have energy for other projects

The Tools Behind the Workflow

Here's the exact stack:

OpenClaw: The orchestration layer. Agents communicate through Telegram, coordinate tasks, access tools.

Obsidian: Shared workspace. Research notes, drafts, and final content all live here.

Substack: Publishing platform. Echo formats for their specific requirements.
>li>Brave Search API: Scout's research engine. Fast, privacy-focused, no rate limits.

What I Learned

1. The human is still the bottleneck. The agents can work 24/7, but I can only review during my waking hours. The constraint is my attention, not their output.

2. Prompts matter more than models. I spent weeks tweaking Scout's research instructions. A well-crafted prompt beats a more expensive model every time.

3. Trust builds over time. In week 1, I double-checked everything. By week 4, I trusted Peter's fact-checking. By week 8, I stopped worrying about Quill's structure.

4. The real value is cognitive load. It's not just 5 hours saved. It's 5 hours of mental energy I can spend on higher-leverage work—like strategy, partnerships, and building new products.

Want to Build Your Own?

This isn't magic. It's just agents, tools, and workflows. If you want to set up your own swarm, I documented the entire process:

OpenClaw Setup Guide — Step-by-step installation and configuration

ClawWork Case Study — How agents generated $15K in 11 hours

The Bottom Line

AI agents aren't coming. They're here. And they're not replacing writers—they're making writers more productive.

The newsletter you're reading right now? Scout found the sources. Quill structured the arguments. Peter verified the facts. I added the voice and hit publish.

This is the future of knowledge work. Not humans vs. AI. Humans with AI. And honestly? It's pretty great.