DEV Community

AI Operator
AI Operator

Posted on

How I Use Claude Code to Run My Entire Content Business Solo (The Full Setup)

#ai #productivity #automation #claudeai

I've been running a content and affiliate site solo for over a year. At some point the manual work got ridiculous: keyword research, writing, editing, image creation, publishing, distribution, community engagement, all of it daily.

I automated most of it using Claude Code and a set of personal AI skills I built. Here is the exact setup.

The Core Architecture

Everything runs from a single repo in ~/.claude. It has:

  • A skills/ directory with SKILL.md files that define reusable workflows
  • An agents/ directory for state files (what ran, when, what got posted where)
  • Hook scripts that inject context before and after every session
  • A daily briefing system that surfaces what needs to run today

The result: I open Claude Code in the morning, see a prioritized list of tasks, and execute them. Most of the "thinking" work is already captured in the skill files.

The Writing Pipeline

My content pipeline chains six steps:

  1. Keyword picker - queries Google Search Console for position 5-20 keywords I can move with fresh content
  2. Competitor research - fetches top-ranking pages, loads them into NotebookLM, extracts a research brief
  3. Article writer - writes a 2,000-3,000 word SEO article using the brief, with internal links and a comparison table
  4. Auditor - checks for banned phrases, missing structure elements, and quality issues, then auto-fixes what it can
  5. Image generator - FLUX.1 for the header image, SVG diagrams for key sections
  6. Publisher - WordPress REST API, then a Twitter thread and Hacker News submission

The whole chain runs in about 90 minutes with minimal intervention. The hard part was not building the tools. It was capturing my own editorial judgment in SKILL.md files that the agent actually follows.

The Community Side

Reddit and forum engagement runs separately:

  • A warmup account on Reddit posts pure value comments for the first 14 days, no links
  • Once the account has 20+ posts, it starts soft-mentioning relevant resources
  • A subreddit I manage gets daily original content, with promotional posts spaced every 5-6 posts
  • Forum outreach runs every two days across niche communities in English, Spanish, and Portuguese

All of this is tracked in state files so the agent knows what's been done and avoids repetition.

What Actually Moves Revenue

The affiliate revenue comes from a few things:

  1. Articles that rank for product-specific keywords with genuine comparison content
  2. Email subscribers who come through lead magnets and get a 5-day mini course
  3. Community members who see consistent value before ever seeing a product mention

The automation handles the volume. The judgment calls about which topics to pursue and which products to promote still require human input, but that is maybe 20 minutes a day.

The Tools That Matter

For anyone building a similar setup, the tools that made the biggest difference:

  • Claude Code with MCP servers (Google Search Console, Gmail, Playwright for browser automation)
  • NotebookLM for research synthesis before writing
  • fal.ai for image generation
  • Systeme.io for email sequences and lead magnet delivery
  • WordPress REST API for headless publishing

The Full Playbook

I documented all of this in a set of guides for people who want to build similar systems without spending months figuring out the architecture:

All at payhip.com/AIOperator.

The biggest lesson from building this: the bottleneck is not tools or models. It is capturing good decisions in a format an agent can follow reliably. Once you solve that, the automation mostly takes care of itself.

Happy to answer questions about any part of the setup.

Top comments (0)