Experiments

I shipped Mintroot — a real Android finance app with 10 entities, 12 milestones, and a 4-layer matching engine — in 48 hours via a single Opus 4.7 1M-context conversation. The 1M context was not the trick. What I did before the first line of Kotlin was.

How a single Go binary turned MCP from a context-burning liability into a zero-overhead CLI powerhouse with Unix composability, data filtering, and security policies

AI coding tools are converging on two outputs -- React + TypeScript when a build pipeline exists, plain HTML when it doesn't. The forces behind this are mechanical, self-reinforcing, and quietly killing language innovation.

How MCPX transforms MCP servers into zero-overhead CLI tools, and why the OpenClaw CEO is right that CLI is the natural interface for agentic systems.

Why structured AI-assisted development with SPARC (Specification, Pseudocode, Architecture, Refinement, Completion) produces 4x better results than raw prompting, and how to adopt it today.

How Tessl pioneered spec-driven development to eliminate AI hallucinations, reduce rework, and make specifications — not code — the source of truth.

How a single Rust CLI proxy compresses command output before it hits the LLM context window, and why pairing RTK with Serena MCP creates the most token-efficient AI coding setup I have ever used.

Why vector search fails at code understanding and how Serena MCP uses the Language Server Protocol to give AI agents the same precise, deterministic vision developers have in their IDEs.

Why full-file rewrites cost 10x more tokens and how symbolic operations with AST-based editing transform AI agent workflows from brute-force to surgical precision.

How I leveraged Claude Code, Gemini, and AI-driven development to build a production-ready portfolio from scratch with zero human code modifications—only planning