Blog

Deep dives on AI agents, Bun/TypeScript architecture, and building for scale.

SUBSCRIBE ON SUBSTACK

I Built a Model Context Protocol (MCP) Server for Obsidian — and It Changed How I Structure AI Workflows

How I turned Obsidian into a structured context layer for AI systems using MCP, and why context architecture matters more than model scale.

WASP-LANG: ORCHESTRATION > BOILERPLATE

Stop rebuilding auth and jobs. Wasp-lang defines architecture declaratively. My evaluation: Speed, AI-readiness, and the DSL trade-off.

AI PRODUCTION: 5 LESSONS FROM THE TRENCHES

Demos are easy. Production is hard. Lessons from Anaqio: Prompt-as-code, Generative Error Handling, and Boring State Machines.

RAG IS DEAD: THE RISE OF CONTEXT DENSITY

Vector search is a bottleneck. I moved to Context-Augmented Generation (CAG). Large windows + Prompt Caching = Superior Reasoning.

AGENTIC NAVIGATION: BEYOND THE RAG CEILING

RAG is a search bar. Agents need navigation. 3 patterns: Tool-first design, Schema-aware retrieval, and Stateful reflection.

BUN MONOREPOS: 45s CI, ZERO TYPE DRIFT

I built Anaqio on Bun Workspaces. No config tax. Next.js 16 + 11 shared packages. Faster iteration, harder types, zero fluff.