I Automated My Entire Blog with AI. It Was a Disaster (At First).

Short summary

Automated a Next.js/Vercel blog using open-source AI agent Hermes for research, writing, image generation, scheduling, and deployment to production. Encountered six critical failures including AI style-rule violations, git config mismatches breaking deployments, API quota limits, timezone bugs in scheduling, concurrent git conflicts, and writing quality issues. Solved each with specific fixes: mandatory self-audit before publishing, 51-word style ban list, proper git config, User-Agent headers, UTC timestamps, and git pull --rebase on all pushes; workflow now reliably produces 85% AI-written content with 15% human refinement.

• •Built automated blog pipeline with Hermes AI agent for research, writing, image generation, scheduling, and Vercel deployment
• •Discovered and debugged six systematic failures: git config mismatches, API quota limits, timezone bugs, concurrent git conflicts, style rule violations, and image generation issues
• •Implemented quality controls including mandatory self-audit steps, 51-word style ban list with tier system, and deployment safeguards resulting in stable 85% AI / 15% human content mix

Generated with AI, which can make mistakes.