Your CLI Isn't Agent-Friendly. Here's What I Fixed.
I rebuilt a 200-command CLI for AI agent consumption. Not the commands themselves, but the help text, output formats, exit codes, and examples. That was enough.
Blog
I stopped fine-tuning and put personality in markdown
Seven years of training AI personality into model weights. Then I wrote it down in a file and got better results. Configuration beat training.
What an AI agent actually did in my homelab rebuild
A full infrastructure rebuild across 35 devices, with Claude Code handling planning, config generation, and many of the execution steps. What worked, what broke, and what still needed a human.
Writing Is Easy. Blogging Is Hard.
I've replatformed four times in four months. The words were never the problem.
The AI Skills Audit
What happens when you treat your AI agent's skill library like production code, and what breaks when you don't.
Plans Are the Engineering
Four projects, 150 plans, and the realization that the plans matter more than the code they produce.
Skippy Gets a Brain
Fine-tuning GPT-3 gave Skippy real personality. Fine-tuning GPT-4o made him insufferable. The transfer learning era of a homebrew chatbot.
Teaching Skippy to Talk
From char-rnn profanity to DialoGPT conversations. How transfer learning turned an angry chatbot into something people actually wanted to talk to.
Skippy Is Born
RNNs, seq2seq, and a Reddit-trained chatbot that came into the world swearing. The moment the project got a name.
RNN Fever Dream
Seq2seq papers land, Torch7 exists, and suddenly my homelab is generating conversations about the meaning of life. The moment AI stopped being a toy.
Markov Chains and Bad Ideas at 3am
Building IRC and Twitter bots with Markov chains, Python, and a homelab that ran hot enough to heat the room. The first experiments that started everything.