I built a small content registry that scans every blog and verifies each post is actually reachable and in the sitemap, not just marked published in a row somewhere.
My scheduled jobs silently failed for two weeks because a token went empty. The fix was to run them as native k3s CronJobs on the node where the runner already lived.
A recipe-driven Playwright engine that submits one product to many directory sites by modeling the form pattern, not the site, with explicit success detection.
How I turned a script into a finished video with code instead of a timeline app, and why determinism is the real win for a solo dev.
How to orchestrate a large generation or audit job across parallel AI agents with adversarial verification and looping quality gates, instead of trusting one giant prompt.
Agentic AI coding gives real leverage only when you decompose work, isolate it, anchor it with durable context, and verify the output. Here is how I direct it.
An honest breakdown of what it costs to build a useful AI agent for a business, the build cost versus the ongoing run cost, and what drives each number.