Running a Newsletter Business With a Team of AI Agents (Full Build)

Key Takeaways

• The architecture: a multi-agent team — research, content, social, and analytics agents, plus a built-in coder agent — that populates an operations dashboard with stages: research → draft → review → schedule → publish (to ConvertKit).
• Agents collaborate in teams. The platform orchestrates sub-agents that delegate to and communicate with each other; you tag a specific agent for a specific task. It runs a 24/7 virtual desktop you can drive from Telegram (or Discord/WeChat).
• Per-agent configuration: each agent gets instructions, tool access, plugins (LinkedIn, X for live analytics), a choice of skills, and its own model family (Gemini / OpenAI / Claude) — all within the subscription.
• Built by prompting. The coder agent assembled the dashboard from natural-language prompts in about 1.5 hours, creating a directory of markdown files and folders so context is retained between runs.
• 24/7 = scheduled tasks. Research Monday 9am, weekly draft Tuesday, social post Thursday, analytics Friday, plus a daily 8am check-in — each runs on repeat and updates the dashboard automatically.
• No connector? Script it. Because there was no built-in ConvertKit connector, he had the agent write code to pull data via API key, then scheduled it to refresh automatically. A human still approves each post and provides editorial judgment.
• Cost framing: a ~$19/month plan (most credits consumed on the one-time build) versus the ~$1,500/month he previously paid a contractor for the same work — with the caveat that quality still needs human review and revisions.