What Happened
A founder posted a detailed breakdown on Hacker News showing how they built a four-agent content pipeline that handles the full cycle from research to publication for $130 per month. The system replaced workflow that previously required a marketing team costing around $200,000 annually.
The stack uses four specialized agents:
- Research Agent ($8/month) - Monitors 50+ RSS feeds and 10 competitor blogs, runs nightly search queries, and scores topic opportunities by search volume growth, competition gaps, and expertise alignment. Outputs 5-7 viable content angles with citations.
- Writer Agent ($25/month) - Uses a two-pass system (outline, then expansion), pulls in brand voice via system prompts, and generates 1,500-2,500 word articles with automatic tool references.
- QA Agent ($12/month) - Enforces a minimum of three citations per piece, checks tone compliance, and evaluates readability using Flesch-Kincaid scoring (Grade 12 ceiling). Auto-approves content that meets thresholds.
- Publisher Agent ($5/month) - Pushes content to the production database, schedules publication 12-36 hours ahead, and logs final URLs.
Total costs break down to $85/month in API calls (Anthropic + OpenAI), $15 for VPS hosting, and $30 for search and scraper APIs.
The results: content output went from 120 articles in Q1 2025 to 487 pieces across channels in Q1 2026, with the full pipeline completing in six hours versus the previous 2-3 week turnaround.
Why It Matters
This is a concrete example of what multi-agent workflows look like in production, not a demo or a concept deck. The numbers tell a clear story: a 4x increase in output at roughly 0.065% of the cost of a human team.
For solo operators and small teams, this kind of architecture is becoming table stakes. If your competitor can publish 487 pieces a quarter while you are manually writing 30, the content gap compounds fast. The $130/month price point also removes the financial barrier entirely. That is less than most individual SaaS subscriptions.
The six-hour cycle time is just as significant as the cost savings. Research-to-publish in six hours means you can respond to trending topics the same day they surface, something that was impossible with a weekly editorial calendar.
Our Take
The architecture here is sound, but the post buries the real question: what is the quality of 487 AI-generated articles? Volume without quality is just spam with better infrastructure. The QA agent checking Flesch-Kincaid and citation counts is a start, but those are mechanical checks. They will not catch bland analysis, surface-level insights, or the kind of generic AI slop that readers skip past.
That said, the modular agent design is worth studying. Separating research from writing from QA means each agent can be swapped, upgraded, or fine-tuned independently. When a better model drops, you upgrade one agent instead of rebuilding everything.
The practical takeaway: you do not need to replicate this exact stack. But if you are still writing every piece from scratch without at least a research agent feeding you scored topic opportunities, you are leaving the most tedious part of the workflow on your own plate. Start with the research agent. That $8/month component probably delivers 80% of the time savings.