AI Lab
My AI Agent Wrote Nine Courses From Nothing. I Published All of Them.
A bypass route, a --force-route=modernize flag, a surviving_words=0 log line nobody read. The bug that turned my LMS into a fabrication factory.
2026-04-298 min read
Blog
Playbooks and practical lessons from shipping games.
AI Lab
How to Use Claude Code Without Drowning in Context
Claude Code works best when you control task framing, session boundaries, and approvals. This practical field guide shows how to keep it sharp instead of overloaded.
2026-04-198 min read
AI Lab
Codex vs Claude Code for Real Software Work (April 2026 Field Report)
A current practitioner view of Codex vs Claude Code based on workflow fit, approvals, scripting, sessions, and review posture in real software work.
2026-04-1910 min read
AI Lab
Why Harness Engineering Matters More Than Prompts
Prompts matter, but they are only one layer of a reliable AI workflow. Here is why harness design, permissions, context control, and review loops matter more.
2026-04-198 min read
AI Lab
My Model-Routing Stack for Real Work
The best AI workflow is usually a routed system, not a single favorite model. Here is a practical framework for assigning the right model to each stage of work.
2026-04-1910 min read