I'm Billy, a BMX rider who spends my weekends hitting rails and weekdays coding up Python. I built an AI agent called OpenClaw to run three of my brands—my gear shop, custom bike frames, and skate-culture newsletter—while I sleep. Here’s how it works.
Tools and Setup
I needed something that could handle the grind of social media posts without me staying awake all night. The system runs 12 posts per day across Instagram, TikTok, Twitter, and YouTube Shorts. I use n8n to orchestrate workflows, Claude for captions, Gemini for images, and OpenClaw as the glue.
The workflow starts at 2 a.m. when n8n triggers a Lambda function that sends a prompt to Claude. The prompt asks Claude to write a 150-character caption with two emojis, a call-to-action, and a relevant hashtag. Claude spits out a draft in seconds.
Next up is image selection using Gemini. I feed it photos from my phone’s camera roll, and Gemini scores them for brightness, composition, and relevance. Anything above 7.5 gets auto-selected; lower scores are queued for manual review. About 60% of the images make the cut automatically. Once selected, the workflow calls each platform's API to upload media, attach captions, and schedule posts for morning.
Content Strategy
All this happens in a single OpenClaw container running on a cheap VPS. The container keeps state for each brand, tracks schedules, and logs performance metrics with SQLite. When a post goes live, it gets logged, and analytics nodes query Instagram Insights, TikTok Analytics, and YouTube Studio for engagement data. This feeds back into Claude's prompts to tweak styles that boost click-through rates.
Since I started running OpenClaw, my follower count has grown by 48% in six months. Average engagement per post shot up from 150 to 2,100 reactions. Time spent on content creation dropped from 30 hours a week to under three.
If you’re into building your own AI agents for social media, check out OpenClaw at axon.nepa-ai.com.