Automating Dev Blog Publishing with Local AI Agents

Description and use

The automation increased publication frequency from 1-2 to ~15 articles per month, tripling total monthly views (600 to 1,800) while reducing author time investment by 95%. However, average views per article dropped, engagement remained flat, and occasional factual hallucinations require manual review.

Problem

Manual blog writing is time-consuming (4-6 hours per article), leading to low publication frequency (1-2 articles/month) and inconsistent presence, despite having ideas and opinions to share.

Outcome

The automation increased publication frequency from 1-2 to ~15 articles per month, tripling total monthly views (600 to 1,800) while reducing author time investment by 95%. However, average views per article dropped, engagement remained flat, and occasional factual hallucinations require manual review.

Workflow

• Schedule a cron job to run every 48 hours.
• Trigger the AI agent (Celebi) via command line.
• Agent reads personal style guide (accounts.md) to ensure tone consistency.
• Agent scans a local draft folder for unpublished ideas or generates new topics based on recent projects.
• Agent uses local LLMs (Qwen 3.5 9B, Qwen 3 Coder 30B, Granite 3.2 8B) to outline, draft, and self-edit the article.
• Agent publishes the final markdown article to Dev.to via REST API.
• Author performs a quick 2-minute skim for factual accuracy before or after publication.

Evidence

I went from 12 articles per year to ~15 per month.

Source

Total monthly writing time: 8-12 hours vs ~30 minutes

Source

Total monthly views: 600 vs 1,800

Source

I caught it claiming that Ollama supports hot-swapping models (it doesn't)

Source