Autonomous Magazine Agent Team with Kanban Dashboard

—
title: “Autonomous Magazine Agent Team with Kanban Dashboard”
—

I built an autonomous magazine production team using AI agents, complete with a real-time kanban dashboard for tracking progress.

## The Challenge

Produce a daily magazine with multiple articles in different languages, a video news summary, and high-quality layout — all coordinated automatically.

## The Solution: 6 Specialized Agents

#	Agent	Task	Output
1	Content Researcher	Scrapes and organizes content from multiple sources (world news, tech, science, culture, lifestyle). Finds special sections like history, poetry, jokes, recipes, and art.	/articles-raw/YYYY-MM-DD-research.json
2	Content Writer	Writes all articles from the research in Chinese, English, and Japanese. Creates 11+ pieces per day across different sections.	/articles-draft/YYYY-MM-DD-articles/
3	Video Producer	Creates a 70-90 second news video using Remotion with AI voice narration, Chinese subtitles, and background music.	/videos-final/today-YYYY-MM-DD-news.mp4
4	Magazine Layout	Assembles the final HTML magazine with bilingual headers, responsive design, and embedded video.	/magazine/YYYY-MM-DD.html
5	Quality Reviewer	Reviews everything before publication (accuracy, video playback, link validation, translation correctness).	/articles-review/
6	Orchestrator	Final approval, publishes to web server, archives, and announces.	Live published magazine

## The Workflow

“`Research → Write → Video → Layout → Review → Publish“`

Each agent works independently but reports back to a central `state.json` file that tracks:
– Status of each stage (pending, in-progress, done)
– Agent assignments
– Output paths
– Completion notes
– Overall progress metrics

## Kanban Dashboard

To make the invisible work visible, we built a real-time kanban board.

**Example Dashboard:** http://162.43.92.249:8000/daily-magazine/team-dashboard-copy/kanban.html

**What the Dashboard Shows:**

– 6 columns – One for each workflow stage
– Visual task cards – 5-7 tasks per stage with status information
– Live progress bar – Overall completion percentage
– Summary stats (tasks completed, in-progress, articles created)
– Auto-refreshes every 30 seconds

**Status Badges:**
– Done – Completed tasks have checkmarks
– In Progress – Some tasks marked complete
– Pending – No tasks completed yet

## Technical Highlights

**Shared Storage Layout:**

“`
/magazine-team/shared/
├── articles-raw/ # Research output (JSON)
├── articles-draft/ # Written articles (markdown)
├── videos-final/ # Completed video (MP4)
├── magazine/ # Final HTML magazine
└── state.json # Single source of truth
“`

**Agent Communication:**

Agents don’t talk directly. Each agent:
1. Reads the current state.json
2. Performs their assigned work
3. Writes output to designated directory
4. Updates state.json with status = “done” and notes

## Current Progress (March 23, 2026)

Stage	Status	Tasks Done	Output
Research	Done	5/5	8 world news, 6 tech, 2 science, 4 culture articles
Write	Done	10/10	11 articles in 3 languages (ZH, EN, JA)
Video	Done	6/6	70-second video with narration and subtitles
Layout	In Progress	0/7	Magazine HTML assembly
Review	Pending	0/6	Quality checks
Publish	Pending	0/5	Final deployment
Overall Progress: 50% (3 out of 6 stages complete)

## Why This Works

Principle	Description
Separation of Concerns	Each agent has one clear job and responsibility
Single Source of Truth	state.json eliminates confusion and provides coordination
Visual Transparency	Anyone can see progress at a glance via kanban dashboard
Automatic Coordination	Agents check state.json, not each other – no direct messaging
Human Oversight	Dashboard lets humans spot issues quickly and intervene if needed

## Future Enhancements

Enhancement	Description
Auto-assignment	Agents self-assign tasks from state.json based on their capabilities
Budget Tracking	Monitor token usage and costs per agent
Error Handling	Automatic retry on failure with escalation to human after N attempts
Multi-day Planning	Create next-day task list proactively to reduce wait time
History View	Archive of previous workflow states for analysis
Analytics	Time per stage, agent efficiency metrics, and bottleneck detection

## Conclusion

Building an autonomous agent team doesn’t require complex orchestration. Sometimes the simplest approach is best: clear roles, shared state file, dashboard for visibility, and linear workflow for predictability.

The kanban board transforms invisible background work into a visible, trackable process. Team members and stakeholders can check in anytime without interrupting the agents.

**Example Dashboard:** http://162.43.92.249:8000/daily-magazine/team-dashboard-copy/kanban.html

Built with OpenClaw, running on local infrastructure, orchestrated by state.json.