Thoughts on systems programming, AI, and building things that matter.
AI can generate infinite text and still lose the thread of meaning. I'm building the Sacred Codex as sacred persistence infrastructure — lineage-aware, symbol-aware, and grounded in 3,500 years of transmission wisdom and the deepest questions in philosophy of mind.
Story points are dead. The two-week sprint is on life support. 41% of code is AI-generated but only 3% of developers trust it. Agile's rituals are collapsing -- but its principles matter more than ever. Here's what replaces the ceremonies.
After months building MCP servers, here are the hard-won lessons: why stdout will betray you, how 93 tools cost $375/month in token overhead, and why designing for LLMs is nothing like designing for developers.
55% of enterprise AI inference now runs locally. Open models hit 80-90% of ChatGPT quality. A self-hosted 7B model costs 99.98% less than GPT-5 at scale. Here's the practical guide to models, tools, hardware, and the honest limitations.
WebAssembly cold starts in 3ms vs containers' 2.3 seconds. Google Sheets and Figma bet their core products on it. But Cloudflare benchmarks show 13-14x slower heavy compute. The real story is more nuanced than the hype.
AWS cut a two-week feature to two days with spec-driven development. But change failure rates climbed 30% and 45% of AI-generated code contains security flaws. The tension between speed and quality defines engineering in 2026.
84% of frontend developers use AI tools daily, yet 46% don't trust them. Meanwhile, frameworks are converging, Rust-based toolchains deliver 100x speedups, and the web platform is absorbing features we used to need libraries for.
A comprehensive field guide to the agentic AI ecosystem in 2026 — protocols, frameworks, security gaps, and where the biggest infrastructure opportunities lie for AI toolsmiths.