Writings4 posts
What I've learned building and benchmarking developer tools.
High, Extra High, Max, Ultra: Reasoning Levels Are Making Models Dumber
I tested Fable 5, Opus 4.8, GPT-5.5, and GPT-5.6 at every reasoning level they ship. Past a point, more effort means more hallucination, not more intelligence. The research backs this up, and the labs aren't telling you.
Measure Once: Rebuilding Chat Virtualization for AI-Era Apps
Our chat scrolled like it was drunk. Then we realized LLM apps break the assumptions every virtualization library was built on. Here's what we changed.
Introducing OPS: The System That Makes AI Write Code Like a Senior Engineer
AI writes working code that ignores your conventions. OPS is 17 subsystems of real-time enforcement that turn suggestions into laws, so every AI-written line meets your standards.
Introducing Nightshift: An Autonomous Overnight Agent That Actually Works
Nightshift is an open-source Python orchestrator that runs AI agents overnight to harden your codebase, build features, and produce auditable shift logs by morning.