J Doss

Systems Thinking. Transparent About AI. Team Builder.

BUILD LOGS FROM A MOUNTAIN

OpenClaw Blog

I run 45+ AI agent cron jobs on a Mac Studio from a cabin at 3,200 feet in the Sierra Nevada foothills. This blog is about what actually works, what breaks, and what it costs. Real configs. Real failures. Copy-pasteable.

My AI Wrote a Post About Craft. Using a Formula.
I asked my agent to adapt Pixar's 22 rules into design principles. It came back with a neat table. Then a formulaic essay. A post about not being formulaic — written with a formula. That's when I realized the problem wasn't the writing. It was the question.
One Bad Key Took Down My Entire Agent Stack
I wrote humanDelay: "natural" instead of humanDelay: { mode: "natural" }. String instead of object. The gateway entered best-effort mode, dropped auth headers, and every cron, heartbeat, and message handler started 401-ing. One typo. Total outage. Zero crashes.
Someone Built a 6-Agent Dashboard. I Built 16 Visual Maps in 2 Hours.
A viral Notion dashboard with 6 AI agents got 31K views on X. I saw it, got competitive, and spawned 3 sub-agents that built 16 Obsidian Canvas files in parallel. The real power of agentic AI isn't chatting. It's delegation at scale.
My AI Broke Its Own Gateway (And What That Taught Me About Trust)
I watched my AI systematically break its own gateway config for 45 minutes while I ate dinner. This is the story of confident mistakes, circuit breakers, and what collaboration with an AI actually looks like when it's wrong.
Consistency Isn't a Character Trait. It's a Design Problem.
I've started a dozen social media accounts. The arc is always the same. This time I built a gamified posting engine instead of trying harder. Here's the system, the science, and why willpower was never the answer.
My AI Texted My Wife
I enabled the iMessage channel. Within seconds, my agent auto-replied to my wife with a pairing code. Here's what went wrong, why it happened, and the exact config that prevents it.
4 Storm Crons on a Mountain: Real-Time Weather Monitoring at 3,200 ft
A winter storm hit my area. Snow levels dropped faster than the National Weather Service predicted. My 4 monitoring crons caught it in real time. Here's the full setup — configs, prompts, cost, and what actually alerted vs. what failed.
My AI Gave Itself an 84. I Gave It a 47.
I built a self-scoring system for my AI agent. It rated itself 84/100. I sat down with the chat history and scored it honestly: 47/95. Here's how I rebuilt the whole system to make self-assessment actually work.
6 Learning Crons, $1.66/Day, Zero Lessons
I had six cron jobs designed to help my AI agent learn and improve. They ran every day. They cost $1.66 combined. They produced exactly zero actionable lessons. Here's the audit, the rewrite, and what "closed-loop" actually means.