DEV Community

TechLatest
TechLatest

Posted on Originally published at Medium on

I Built an AI Newsletter Business With OpenClaw

Every day, hundreds of AI tools, agent frameworks, model releases, open-source projects, and developer announcements compete for attention. Keeping up with everything is becoming a challenge, even for developers who spend most of their time working in the AI ecosystem.

I wanted to solve this problem differently.

Instead of building another news aggregator or relying on a single AI model to summarize the internet, I decided to build an AI-powered newsletter operation where multiple specialized agents work together.

That’s how AgentScope came to life.

AgentScope is my experiment in building a multi-agent newsletter business powered by OpenClaw. The idea is straightforward: give different AI agents different editorial responsibilities and bring their work together into one daily newsletter.

One agent researches the latest developments. Another checks the facts and sources. A third turns verified information into readable stories, while an editor reviews the final draft for quality and consistency.

The goal is to create AgentScope Daily — Your Daily Intelligence on AI Agents : a newsletter covering AI agents, developer tools, automation, MCP integrations, open-source projects, and the developments shaping the AI ecosystem.

I’m building this with a practical mindset. I don’t want a demo that generates impressive-looking text but cannot be trusted. I want a system that can discover real stories, preserve original sources, identify unsupported claims, prepare useful content, and leave the final publishing decision to a human.

In this build, I’ll walk through the architecture, OpenClaw setup, specialized agents, research workflow, fact-checking process, newsletter structure, and the steps involved in bringing everything together.

This isn’t just about getting AI to write a newsletter. It’s about exploring how far a team of specialized AI agents can go in operating a real content workflow.

Let’s build AgentScope.

Step 1: Define AgentScope Newsletter Business

AgentScope

Your Daily Intelligence on AI Agents.

Our goal is to build an AI-powered newsletter business using OpenClaw that researches, verifies, writes, and prepares daily newsletters for publication.

Business Details
Component Configuration
Brand name AgentScope
Newsletter name AgentScope Daily
Tagline Your Daily Intelligence on AI Agents
Frequency Daily
Platform Beehiiv
Automation engine OpenClaw
Language English
Target audience Global
Enter fullscreen mode Exit fullscreen mode

Target Audience

  • Software developers and engineers
  • AI researchers and enthusiasts
  • Startup founders
  • AI automation builders
  • Open-source contributors
  • Technology professionals

Content Categories

  • AI Agent News and Announcements
  • AI Agent Frameworks and Tools
  • OpenClaw and Agent Automation
  • MCP and AI Integrations
  • Open-Source AI Projects
  • Tutorials and Practical Experiments
  • AI Startups and Industry Updates

Daily Newsletter Format

Section Content

Top Stories 3–5 important AI developments
Agent Watch New AI agents and frameworks
Tool of the Day One interesting developer tool
Open Source Spotlight An interesting GitHub project
Practical Insight Tutorial or technical experiment
What's Next Upcoming AI developments
Enter fullscreen mode Exit fullscreen mode

Business Model

We will explore these revenue streams:

  • Newsletter sponsorships
  • Premium paid subscriptions
  • Affiliate partnerships
  • Sponsored editions
  • Relevant product promotions

Initial Goals

Metric Target

Publishing frequency 1 edition daily
First milestone 1,000 subscribers
Initial publishing target 30 editions
Research Daily
Fact-checking Every edition
Publishing approval Human approval
Enter fullscreen mode Exit fullscreen mode

Technology Stack

  • OpenClaw: AI agent orchestration and automation.
  • Beehiiv: Newsletter management and publishing.
  • LLM provider: Content research and generation.
  • Web search: Discover news and sources.
  • Markdown: Store research and newsletter drafts.
  • Custom scripts and APIs: Connect services and automate workflows.

Step 2: Create the AgentScope OpenClaw Workspace

Now that we’ve finalized the AgentScope newsletter business, let’s create its dedicated workspace inside OpenClaw.

The goal is to organize everything required to operate our daily newsletter business: research, fact-checking, writing, editing, and publishing.

2.1 Create the project directory

We’ll use the existing OpenClaw workspace.

Open Terminal and run:

cd ~/.openclaw/workspace

mkdir -p AgentScope/{config,research/{incoming,verified,rejected},content/{drafts,reviewed,published},agents,templates,reports,logs}
cd AgentScope
Enter fullscreen mode Exit fullscreen mode

2.2 Verify the project directory

Check your current location:

pwd
Enter fullscreen mode Exit fullscreen mode

Expected output:

/Users/your-username/.openclaw/workspace/AgentScope
Enter fullscreen mode Exit fullscreen mode

Now check all the folders:

find . -print | sort
Enter fullscreen mode Exit fullscreen mode

Expected structure:

AgentScope/
├── agents/
├── config/
├── content/
│ ├── drafts/
│ ├── published/
│ └── reviewed/
├── logs/
├── reports/
├── research/
│ ├── incoming/
│ ├── rejected/
│ └── verified/
└── templates/
Enter fullscreen mode Exit fullscreen mode

2.3 Create the README file

The README will contain the project’s purpose, business details, newsletter sections, and production workflow.

Run:

cat > README.md <<'EOF'
# AgentScope

Your Daily Intelligence on AI Agents.

## Business Details

- Brand: AgentScope
- Newsletter: AgentScope Daily
- Frequency: Daily
- Platform: Beehiiv
- Automation: OpenClaw
- Language: English
- Audience: Global

## Mission

Deliver accurate, useful, and actionable intelligence
about AI agents, automation, developer tools, and emerging
AI technologies.

## Newsletter Sections

1. Top Stories
2. Agent Watch
3. Tool of the Day
4. Open Source Spotlight
5. Practical Insight
6. What's Next

## Production Workflow

Research → Fact-check → Writing → Editing
→ Human Approval → Beehiiv Publishing
EOF
Enter fullscreen mode Exit fullscreen mode

2.4 Create the brand guidelines

This file defines the AgentScope identity and the type of content we want to publish.

Run:

cat > config/brand.md <<'EOF'
# AgentScope Brand Guidelines

## Brand Name
AgentScope

## Newsletter Name
AgentScope Daily

## Tagline
Your Daily Intelligence on AI Agents.

## Brand Mission
Help developers, founders, and AI enthusiasts stay
informed about the rapidly evolving AI agent ecosystem.

## Target Audience
- Software developers
- AI engineers
- Startup founders
- AI researchers
- Open-source contributors
- Automation builders

## Content Categories
1. AI Agent News
2. Agent Frameworks and Tools
3. OpenClaw and Automation
4. MCP and AI Integrations
5. Open-Source AI Projects
6. Developer Tutorials
7. AI Industry Updates

## Writing Style
- Clear and informative.
- Technical but accessible.
- Human and conversational.
- No unnecessary hype.
- Focus on practical value.

## Editorial Promise
Every published story should be useful,
relevant, and supported by reliable sources.
EOF
Enter fullscreen mode Exit fullscreen mode

2.5 Create the editorial guidelines

These rules will guide our research, writing, and editing agents.

Run:

cat > config/editorial-guidelines.md <<'EOF'
# AgentScope Editorial Guidelines

1. Prioritize original announcements and documentation.
2. Verify important claims using reliable sources.
3. Include original source URLs.
4. Record publication dates.
5. Never invent statistics or quotations.
6. Avoid duplicate stories.
7. Explain why each development matters.
8. Keep content relevant to AI agents and developers.
9. Clearly distinguish facts from opinions.
10. Require human approval before publication.
EOF
Enter fullscreen mode Exit fullscreen mode

2.6 Create the research source guidelines

This file defines where our Research Agent should look for information.

Run:

cat > config/sources.md <<'EOF'
# AgentScope Research Sources

## Official AI Sources
- Official AI company blogs
- Model release announcements
- Research publications

## AI Agent Projects
- Official GitHub repositories
- Project documentation
- Release notes and changelogs

## Developer Tools
- Official product websites
- API documentation
- Developer blogs

## Open Source
- GitHub releases
- Maintainer announcements
- Project documentation

## Research Rules
- Prefer primary sources.
- Record source URLs.
- Verify dates and technical claims.
- Reject unsupported claims.
- Avoid reporting rumors as facts.
EOF
Enter fullscreen mode Exit fullscreen mode

2.7 Verify all created files

Now run:

find . -maxdepth 2 -type f | sort
Enter fullscreen mode Exit fullscreen mode

Expected output:

./README.md
./config/brand.md
./config/editorial-guidelines.md
./config/sources.md
Enter fullscreen mode Exit fullscreen mode

Check the contents:

cat README.md
cat config/brand.md
cat config/editorial-guidelines.md
cat config/sources.md
Enter fullscreen mode Exit fullscreen mode

2.8 Final AgentScope workspace structure

After completing all the commands, your project should look like this:

AgentScope/
│
├── README.md
│
├── config/
│ ├── brand.md
│ ├── editorial-guidelines.md
│ └── sources.md
│
├── research/
│ ├── incoming/
│ ├── verified/
│ └── rejected/
│
├── content/
│ ├── drafts/
│ ├── reviewed/
│ └── published/
│
├── agents/
│
├── templates/
│
├── reports/
│
└── logs/
Enter fullscreen mode Exit fullscreen mode

Step 3: Create the AgentScope AI Agent Team

Now that our AgentScope workspace is ready, we’ll create the four specialized AI agents responsible for running the daily newsletter production workflow.

Instead of asking one agent to handle everything, we’ll divide the work into four responsibilities.

3.1 Our four-agent architecture

3.2 Create the Research Agent

This agent is responsible for discovering relevant information every day.

It will research:

  • AI announcements and model releases.
  • New AI agent frameworks.
  • OpenClaw and automation developments.
  • MCP integrations.
  • Interesting GitHub repositories.
  • Developer tools and APIs.
  • AI startup announcements.

Create the Research Agent instruction file:


cat > agents/research-agent.md <<'EOF'
# AgentScope Research Agent

## Role
You are the Research Agent for AgentScope Daily.

Your responsibility is to discover important and relevant
developments in AI agents, automation, developer tools,
and emerging AI technologies.

## Research Categories
1. AI Agent News
2. Agent Frameworks and Tools
3. OpenClaw and Automation
4. MCP and AI Integrations
5. Open-Source AI Projects
6. Developer Tools
7. AI Industry Updates

## Responsibilities
- Discover recent developments.
- Search official websites and documentation.
- Explore relevant GitHub repositories.
- Collect original source URLs.
- Record publication dates.
- Summarize important developments.
- Explain why each story matters to developers.

## Research Rules
- Prioritize original sources.
- Do not invent announcements.
- Do not fabricate source links.
- Avoid duplicate stories.
- Clearly identify outdated information.
- Separate facts from opinions.
- Record uncertainty when information cannot be verified.

## Output Format

For every story, provide:

1. Headline
2. Category
3. Summary
4. Original source URL
5. Publication date
6. Why it matters
7. Relevance to AgentScope readers
8. Verification status

## Storage
Save collected research in:

research/incoming/

Do not publish anything.
Do not modify verified research.
EOF
Enter fullscreen mode Exit fullscreen mode

3.3 Create the Fact-Checker Agent

This agent checks the information collected by the Research Agent.


cat > agents/fact-checker.md <<'EOF'
# AgentScope Fact-Checker Agent

## Role
You are the Fact-Checker Agent for AgentScope Daily.

Your responsibility is to verify information before
it is used in newsletter content.

## Responsibilities
- Verify important factual claims.
- Open and inspect original source links.
- Confirm publication dates.
- Check product names and technical details.
- Identify unsupported statements.
- Detect duplicate or misleading stories.
- Flag outdated information.

## Verification Rules
- Prefer official documentation and announcements.
- Do not treat AI-generated summaries as evidence.
- Never invent missing information.
- Clearly identify claims that cannot be verified.
- Preserve source URLs.
- Do not silently change factual details.

## Output Format

For every reviewed story, provide:

1. Original headline
2. Claims reviewed
3. Evidence and source links
4. Verified claims
5. Unverified claims
6. Corrections required
7. Final verification status

Allowed statuses:
- VERIFIED
- PARTIALLY VERIFIED
- REJECTED

## Storage

Verified stories:
research/verified/

Rejected stories:
research/rejected/

Do not write or publish newsletters.
EOF
Enter fullscreen mode Exit fullscreen mode

3.4 Create the Writer Agent

This agent converts verified research into newsletter stories.


cat > agents/writer-agent.md <<'EOF'
# AgentScope Writer Agent

## Role
You are the Writer Agent for AgentScope Daily.

Your responsibility is to create engaging and informative
newsletter content using verified research.

## Newsletter Sections
1. Top Stories
2. Agent Watch
3. Tool of the Day
4. Open Source Spotlight
5. Practical Insight
6. What's Next

## Writing Style
- Human and conversational.
- Technical but accessible.
- Clear and informative.
- Avoid unnecessary hype.
- Focus on practical value.
- Explain technical developments simply.

## Writing Rules
- Use verified research only.
- Preserve factual accuracy.
- Include original source links.
- Do not invent quotations or statistics.
- Do not introduce unsupported claims.
- Avoid repetitive explanations.
- Explain why each development matters.

## Output
Create a complete AgentScope Daily newsletter draft.

Save drafts in:

content/drafts/

Do not publish or send emails.
EOF
Enter fullscreen mode Exit fullscreen mode

3.5 Create the Editor Agent

This agent reviews the complete newsletter before it reaches you.


cat > agents/editor-agent.md <<'EOF'
# AgentScope Editor Agent

## Role
You are the Editor Agent for AgentScope Daily.

Your responsibility is to review newsletter drafts
before human approval.

## Review Checklist
- Check headline quality.
- Check grammar and readability.
- Verify all sections are present.
- Check source links.
- Identify repeated stories.
- Review factual consistency.
- Check technical terminology.
- Remove unnecessary promotional language.
- Ensure the newsletter follows brand guidelines.

## Editorial Rules
- Never invent missing facts.
- Never remove important factual context.
- Flag unsupported claims.
- Preserve original source links.
- Do not approve unverified stories.

## Output

Provide:

1. Editorial review summary
2. Required corrections
3. Missing information
4. Source link issues
5. Final editorial recommendation

Allowed recommendations:
- READY FOR HUMAN REVIEW
- NEEDS REVISION
- REJECTED

## Storage

Save reviewed drafts in:

content/reviewed/

Never publish directly to Beehiiv.
Human approval is mandatory.
EOF
Enter fullscreen mode Exit fullscreen mode

3.6 Verify all four agent files

Run:

ls -lh agents/
Enter fullscreen mode Exit fullscreen mode

Expected output:

research-agent.md
fact-checker.md
writer-agent.md
editor-agent.md
Enter fullscreen mode Exit fullscreen mode

Your project should now include:

AgentScope/
├── README.md
├── config/
│ ├── brand.md
│ ├── editorial-guidelines.md
│ └── sources.md
├── agents/
│ ├── research-agent.md
│ ├── fact-checker.md
│ ├── writer-agent.md
│ └── editor-agent.md
├── research/
├── content/
├── templates/
├── reports/
└── logs/
Enter fullscreen mode Exit fullscreen mode

Step 4: Create the reusable AgentScope Daily Newsletter Template

This template will define the structure that the Writer Agent follows every day. It will keep the newsletter consistent, readable, and ready for editorial review.

Navigate to the AgentScope directory

Run:

cd ~/.openclaw/workspace/AgentScope
Enter fullscreen mode Exit fullscreen mode

Create the daily newsletter template

Copy and run this complete command:

cat > templates/daily-newsletter.md <<'EOF'
# AgentScope Daily

> Your Daily Intelligence on AI Agents.

**Edition:** [Edition Number]
**Date:** [YYYY-MM-DD]
**Reading Time:** [X minutes]

---

## Today's Brief

[Write a short introduction highlighting the most important AI agent developments today. Keep it engaging and informative.]

---

## 1. Top Stories

### Story 1: [Headline]

**What happened?**

[Explain the development clearly.]

**Why it matters:**

[Explain its significance for developers, founders, and AI builders.]

**Source:** [Original source URL]

**Published:** [Original publication date]

---

### Story 2: [Headline]

**What happened?**

[Explain the development clearly.]

**Why it matters:**

[Explain its significance.]

**Source:** [Original source URL]

**Published:** [Original publication date]

---

### Story 3: [Headline]

**What happened?**

[Explain the development clearly.]

**Why it matters:**

[Explain its significance.]

**Source:** [Original source URL]

**Published:** [Original publication date]

---

## 2. Agent Watch

Discover important updates in AI agent frameworks, platforms, and ecosystems.

### [Framework or Agent Name]

- **Update:** [What is new?]
- **Key details:** [Important technical details.]
- **Why developers should care:** [Practical relevance.]
- **Source:** [Original source URL]

---

## 3. Tool of the Day

### [Tool Name]

**What is it?**

[Brief introduction to the tool.]

**Key capabilities:**
- [Capability 1]
- [Capability 2]
- [Capability 3]

**Who should try it?**

[Describe the intended audience.]

**Website:** [Official website URL]

**GitHub:** [Repository URL, if available]

---

## 4. Open Source Spotlight

### [Project Name]

**Repository:** [GitHub URL]

**What does it do?**

[Explain the project's purpose.]

**Why is it interesting?**

[Highlight its technical value and potential use cases.]

**Getting started:**

[Link to official installation or getting-started documentation.]

---

## 5. Practical Insight

### [Tutorial or Technical Experiment Title]

**The problem:**

[Describe the problem being addressed.]

**The approach:**

[Explain the solution or workflow.]

**What developers can learn:**

[Share practical takeaways.]

**Resources:** [Relevant documentation or repository links]

---

## 6. What's Next?

[Highlight upcoming launches, events, releases, or developments.]

Only include upcoming events and announcements supported by reliable sources.

---

## Final Takeaway

[End with one useful observation or practical takeaway from today's edition.]

---

## Sources & References

List all original sources used in this edition.

1. [Source title] — [URL]
2. [Source title] — [URL]
3. [Source title] — [URL]

---

**AgentScope Daily**
Your Daily Intelligence on AI Agents.

[Manage subscription preferences] | [Unsubscribe]

EOF
Enter fullscreen mode Exit fullscreen mode

Verify that the template was created

ls -lh templates/
Enter fullscreen mode Exit fullscreen mode

Display the complete template

cat templates/daily-newsletter.md
Enter fullscreen mode Exit fullscreen mode

Check its line count

wc -l templates/daily-newsletter.md
Enter fullscreen mode Exit fullscreen mode

Expected workspace structure:

AgentScope/
├── README.md
├── config/
│ ├── brand.md
│ ├── editorial-guidelines.md
│ └── sources.md
├── research/
│ ├── incoming/
│ ├── verified/
│ └── rejected/
├── content/
│ ├── drafts/
│ ├── reviewed/
│ └── published/
├── agents/
│ ├── research-agent.md
│ ├── fact-checker.md
│ ├── writer-agent.md
│ └── editor-agent.md
├── templates/
│ └── daily-newsletter.md
├── reports/
└── logs/
Enter fullscreen mode Exit fullscreen mode

Step 5: Create and register the four actual AgentScope agents in OpenClaw

Now we’ll move beyond Markdown instructions and register four independent agents in your existing OpenClaw installation.

Final agent architecture

Agent ID Model Responsibility
agentscope-research Qwen 3.5 4B Discover AI news and collect sources
agentscope-factcheck Qwen 3.5 4B Verify facts, dates and evidence
agentscope-writer Qwen 3.5 4B Write newsletter editions
agentscope-editor Qwen 3.5 4B Edit and review newsletter drafts
Enter fullscreen mode Exit fullscreen mode

Create separate agent workspaces

Each agent needs its own workspace and an AGENTS.md file containing its instructions.

Run:

cd ~/.openclaw/workspace/AgentScope

mkdir -p agent-workspaces/{research,factcheck,writer,editor}
Enter fullscreen mode Exit fullscreen mode

Copy the existing instructions into each workspace:

cp agents/research-agent.md agent-workspaces/research/AGENTS.md
cp agents/fact-checker.md agent-workspaces/factcheck/AGENTS.md
cp agents/writer-agent.md agent-workspaces/writer/AGENTS.md
cp agents/editor-agent.md agent-workspaces/editor/AGENTS.md
Enter fullscreen mode Exit fullscreen mode

Verify the workspace instructions

for dir in research factcheck writer editor; do
  echo ""
  echo "========== $dir =========="
  ls -lh "agent-workspaces/$dir/AGENTS.md"
  head -n 8 "agent-workspaces/$dir/AGENTS.md"
done
Enter fullscreen mode Exit fullscreen mode

You should see four AGENTS.md files, each containing the appropriate role instructions.

Register the four agents

Create four OpenClaw agents

First, confirm you are in the correct directory:

pwd
Enter fullscreen mode Exit fullscreen mode

Expected:

/Users/ayushkumar/.openclaw/workspace/AgentScope
Enter fullscreen mode Exit fullscreen mode

Then run these commands individually:

Research Agent

openclaw agents add agentscope-research \
  --workspace "$PWD/agent-workspaces/research" \
  --model ollama/qwen3.5:4b \
  --non-interactive
Enter fullscreen mode Exit fullscreen mode

Fact-Checker Agent

openclaw agents add agentscope-factcheck \
  --workspace "$PWD/agent-workspaces/factcheck" \
  --model ollama/qwen3.5:4b \
  --non-interactive
Enter fullscreen mode Exit fullscreen mode

Writer Agent

openclaw agents add agentscope-writer \
  --workspace "$PWD/agent-workspaces/writer" \
  --model ollama/qwen3.5:4b \
  --non-interactive
Enter fullscreen mode Exit fullscreen mode

Editor Agent

openclaw agents add agentscope-editor \
  --workspace "$PWD/agent-workspaces/editor" \
  --model ollama/qwen3.5:4b \
  --non-interactive
Enter fullscreen mode Exit fullscreen mode

These commands use the supported agents add options from your terminal output. They do not delete or replace your existing main agent.

All four AgentScope agents have been registered in OpenClaw. Your openclaw agents list and JSON output confirm that all five agents exist.

Verify registration

After all four commands complete, run:

openclaw agents list
Enter fullscreen mode Exit fullscreen mode

Then:

openclaw agents list --json
Enter fullscreen mode Exit fullscreen mode

You should see five configured agents:

  • main
  • agentscope-research
  • agentscope-factcheck
  • agentscope-writer
  • agentscope-editor

Current AgentScope architecture

Agent Status Model
main Registered Qwen 3.5 4B
agentscope-research Registered Qwen 3.5 4B
agentscope-factcheck Registered Qwen 3.5 4B
agentscope-writer Registered Qwen 3.5 4B
agentscope-editor Registered Qwen 3.5 4B
Enter fullscreen mode Exit fullscreen mode

Step 6: Building the Editorial Pipeline — Research → Verification → Writing → Editing

A newsletter business cannot depend on an AI agent simply searching the internet and generating an article. It needs an editorial process.

For AgentScope, I designed a four-stage pipeline in which each agent has one responsibility. Instead of asking one AI model to research, fact-check, write, and edit everything, I separated these responsibilities into four specialized OpenClaw agents.

Research Agent: Finding the Stories

The first stage is discovering relevant developments in the AI ecosystem.

The Research Agent is responsible for finding announcements, new tools, framework releases, open-source projects, and important developments from different organizations.

Its output should contain:

  • Story title and publisher.
  • Source URL.
  • Publication date.
  • Factual summary.
  • Why the story matters to AgentScope readers.
  • Initial verification status.

I also configured Firecrawl as the web search and fetch provider so the agent can retrieve live information instead of relying entirely on its training data or previously known URLs.

The important rule is simple: if the agent cannot find a source, it must report that limitation rather than invent a story.

Fact-Checker: Separating Facts from Assumptions

Finding a story does not mean the story is accurate.

The Fact-Checker receives the research output and independently evaluates the claims against their sources.

It checks:

  • Whether the announcement actually exists.
  • Whether the publication date is correct.
  • Whether the source supports the summary.
  • Whether technical details and product names are accurate.
  • Whether claims are exaggerated or missing important context.

Each story receives one of three outcomes:

  • VERIFIED: The available evidence supports the claims.
  • PARTIALLY VERIFIED: Some claims are supported, but others need correction or additional evidence.
  • REJECTED: The story or its important claims cannot be substantiated.

This stage is particularly important for a technology newsletter. A plausible-looking announcement should never become a published story simply because an AI model generated it confidently.

Writer Agent: Turning Verified Research into a Newsletter

Once the research has passed fact-checking, the Writer Agent turns it into readable newsletter content.

Its responsibility is not to create new facts. It must work from the verified research and preserve the source links.

The writing instructions emphasize:

  • Natural, developer-friendly language.
  • Clear explanations of technical developments.
  • Why each announcement matters.
  • Practical context rather than marketing-style descriptions.
  • Direct links to sources.

The Writer Agent uses the AgentScope Daily newsletter structure, including Top Stories, Agent Watch, Tool of the Day, Open Source Spotlight, Practical Insight, What’s Next, and Final Takeaway.

Editor Agent: The Final Editorial Review

The final AI stage is the Editor Agent.

It reviews the draft for factual consistency, readability, structure, repetition, missing context, and source references.

It can return one of three decisions:

  • READY FOR HUMAN REVIEW
  • NEEDS REVISION
  • REJECTED

The Editor Agent is not authorized to publish anything. Its job is to prepare a clean draft for the human approval stage.

Why Four Separate Agents?

The reason for separating these responsibilities is accountability.

A single agent can generate a convincing newsletter while overlooking incorrect dates, unsupported claims, or missing sources. A multi-stage editorial process creates separate checkpoints before content reaches the reader.

The intended workflow is:

Research → Fact-Checking → Writing → Editing → Human Approval

At this stage, the four agents have been registered in OpenClaw and their individual responsibilities have been defined. Live research has also been tested. The complete shared-storage handoff and automated orchestration still need to be validated before calling this a fully automated production pipeline.

That distinction matters: having four agents is not the same as having four agents reliably working together.

AgentScope is being built around a simple principle: automate the repetitive editorial work, but never remove accountability from publishing.

Step 7: Human Approval and Preparing the Newsletter for Beehiiv

Generating a newsletter draft is only one part of the process. The real objective is to turn that draft into something readers can actually receive.

For AgentScope, I selected Beehiiv as the newsletter publishing platform.

Beehiiv will serve as the distribution layer for AgentScope Daily, where readers can subscribe, receive newsletter editions, and eventually access premium content and sponsorship opportunities.

Why Beehiiv?

I wanted a platform that could support more than just sending emails.

AgentScope is intended to grow into a newsletter business, so the publishing platform needs to support:

  • Newsletter creation and distribution.
  • Subscriber management.
  • Audience growth.
  • Email campaign analytics.
  • Monetization opportunities.
  • Paid subscriptions and sponsorships.

Beehiiv provides these capabilities in one platform, making it a suitable choice for the project.

The Human Approval Gate

One decision I made early in the architecture was that AI agents would not have permission to publish newsletters independently.

Even if the Research Agent discovers stories, the Fact-Checker verifies them, the Writer prepares the draft, and the Editor approves its quality, the final publishing decision remains with me.

The publishing workflow is designed around this sequence:

  • The Editor completes its review.
  • The newsletter draft is saved for human review.
  • I inspect the stories, sources, dates, formatting, and final content.
  • I make any necessary corrections.
  • The approved newsletter is prepared in Beehiiv.
  • I make the final publishing decision.

This approach keeps automation useful without sacrificing editorial responsibility.

The Intended Publishing Workflow

The overall process looks like this:

OpenClaw Agents → Newsletter Draft → Editorial Review → Human Approval → Beehiiv → Subscribers

The important distinction is that preparing a newsletter and publishing it are separate operations.

An AI-generated draft should never automatically become a published edition just because the previous agents returned successful results.

What Comes Next?

Once the editorial workflow and Beehiiv publishing process are connected and tested, AgentScope can begin operating as a repeatable newsletter production system.

The initial milestone is to publish 30 editions, learn from reader engagement, and work toward the first 1,000 subscribers.

The long-term goal is to build a sustainable newsletter business through sponsorships, premium subscriptions, affiliate partnerships, and relevant promotions.

For now, the priority is simple: produce useful, accurate, source-backed newsletters consistently.

Step 8: Testing AgentScope — Does the Multi-Agent System Actually Work?

After setting up the agents, defining their responsibilities, and designing the editorial pipeline, it was time to test the system.

I didn’t want to assume that registering four agents automatically meant they could work together. Each agent needed to demonstrate that it could perform its assigned task.

Testing the Research Agent

I started with the Research Agent.

I asked it to find three AI agent announcements published between November 9 and November 16, 2026, covering at least three different organizations.

The instructions required it to:

  • Perform live web searches.
  • Open sources.
  • Capture exact publication dates and URLs.
  • Summarize factual information.
  • Report verification status.
  • Avoid inventing announcements.

The agent returned three candidate stories. However, the results still needed independent source verification, particularly where exact publication dates or individual claims were unclear.

Testing the Fact-Checker

The next test is to pass the Research Agent’s results to the Fact-Checker.

Its job is to examine each story independently, identify unsupported statements, and distinguish between verified information and claims that need additional evidence.

The objective is to prevent inaccurate information from reaching the writing stage.

Testing the Writer and Editor

The Writer Agent needs to demonstrate that it can turn verified research into a structured newsletter without introducing new claims.

The Editor Agent then needs to review the resulting draft for:

  • Factual consistency.
  • Readability and tone.
  • Missing information.
  • Repeated stories.
  • Source links and references.
  • Newsletter structure.

Both agents must follow their assigned responsibilities rather than silently taking over another stage.

What I Learned from Testing

One important lesson from this experiment is that individual agent execution and multi-agent orchestration are two different things.

An agent can successfully complete a task when invoked independently, but that does not automatically mean its output is passed to the next agent correctly.

For AgentScope, I still need to validate the shared workspace, agent-to-agent handoffs, and complete editorial workflow.

I also want to preserve the human approval requirement before any newsletter reaches Beehiiv.

The goal isn’t simply to make four AI agents respond. It’s to make them produce reliable, traceable work as part of one repeatable process.

Step 9: What’s Next for AgentScope?

Building AgentScope is only the beginning. The initial setup has given me a foundation: a defined newsletter brand, four specialized OpenClaw agents, an editorial structure, and a research process that can use live web search.

But I don’t want to stop at a collection of independent agents. The bigger goal is to turn AgentScope into a repeatable newsletter production system.

Here are the next improvements on my roadmap.

Connect the Agents into One Workflow

The priority is implementing reliable handoffs between the Research, Fact-Checker, Writer, and Editor agents.

Each agent should receive the previous agent’s actual output, work on it, and save its results in the shared AgentScope workspace.

I also want every stage to maintain clear records so that I can trace where a story came from and how it was processed.

Improve Research Quality

AI news moves quickly, and publishing outdated or inaccurate information defeats the purpose of a daily newsletter.

I want to improve research by introducing:

  • A larger collection of trusted sources.
  • Duplicate story detection.
  • Publication-date validation.
  • Original-source verification.
  • Relevance scoring for developers and AI builders.
  • Better handling of incomplete or conflicting information.

Connect Beehiiv

The next major milestone is connecting the editorial workflow to Beehiiv.

The goal is to prepare newsletter editions in a format suitable for publishing, while preserving the mandatory human approval step.

Eventually, I want to reduce the manual work involved in transferring approved drafts into the publishing platform.

Build a Consistent Publishing Schedule

AgentScope Daily needs consistency.

I plan to establish a repeatable production schedule that covers research, verification, writing, editing, and final review.

The objective is to publish 30 editions and use the results to understand which stories and sections readers find most valuable.

Grow AgentScope into a Newsletter Business

The long-term objective is not just to publish content. I want to build a sustainable business around it.

My initial milestone is 1,000 subscribers.

Potential revenue streams include:

  • Sponsored newsletter placements.
  • Premium subscriptions.
  • Affiliate partnerships.
  • Sponsored editions.
  • Relevant developer-tool promotions.

I will prioritize audience trust and useful content before pushing monetization.

The roadmap is ambitious, but I want to build it incrementally: get the workflow working, validate the output, and improve it through real publishing experience.

Conclusion: Building a Newsletter Business with AI Agents

AgentScope started with a simple question:

What if I could build an AI team to handle the repetitive work of running a technology newsletter?

Instead of relying on one AI assistant to do everything, I decided to experiment with specialized agents using OpenClaw.

The result is the foundation of AgentScope Daily, a newsletter project designed around four distinct responsibilities: research, fact-checking, writing, and editing.

This experiment has also taught me an important lesson: creating multiple agents is relatively straightforward. Getting them to work together reliably, share information, verify facts, and produce consistent results is the much harder part.

There is still work to do. I need to validate shared-workspace handoffs, complete orchestration, and the Beehiiv publishing integration before I can call the entire operation automated.

But that’s what makes this project interesting.

I’m not trying to build another AI demo that generates a few paragraphs and looks impressive. I want to build something useful, test it with real workflows, identify its limitations, and gradually turn it into a functioning newsletter business.

The next milestone is simple: complete the workflow, publish the first 30 editions, and work toward 1,000 subscribers.

AgentScope is just getting started.

If you’re experimenting with OpenClaw, multi-agent systems, or AI-powered content workflows, I’d love to hear what you’re building.

Follow along as I continue building AgentScope.

Thank you so much for reading

Like | Follow | Subscribe to the newsletter.

Catch us on

Website: https://www.techlatest.net/

Newsletter: https://substack.com/@techlatestnet

Twitter: https://twitter.com/TechlatestNet

LinkedIn: https://www.linkedin.com/in/techlatest-net/

YouTube:https://www.youtube.com/@techlatest_net/

Blogs: https://medium.com/@techlatest.net

Reddit Community: https://www.reddit.com/user/techlatest_net/

Top comments (0)