The End of AI Magic.
"I stopped treating LLMs like magic wands and started treating them like CPUs. These are the logs."
The AI industry is drowning in hype, metaphors, and "magic wands." I got tired of it. Every day a new paper drops, a new model ships, and 90% of it is marketing slop.
I'm not here to "surf the AI wave." I'm here to build systems that don't crash when the context window fills up. I spend my days writing specs, enforcing boundaries, and cleaning up the mess left by autonomous agents.
The Boundary Layer
I needed a deterministic boundary layer for my agents. The rule is simple: AI output has to respect technical contracts before it becomes project state.
Studying the Gurus
I don't guess. I read. I spend hours deconstructing Andrej Karpathy's insights on LLM mental models, mapping the "Software 3.0" philosophy. I obsess over the technical clarity of Dan Abramov and the pragmatic engineering of Kent C. Dodds.
I am a student of their rigor. I take their papers and build them until they break.
What you'll find here
- Incident Reports: Raw logs of agent failures and the architectural changes needed to fix them.
- Technical Audits: Deconstructions of trending tools and papers.
- Agent Architecture: The ongoing evolution of deterministic boundaries around probabilistic tools.