A three-pillar workflow for agentic coding: spec-driven planning, multi-LLM review, and hand-curated tests with golden datasets. Each pillar catches a class of failure the other two can't.
Blog
How I modernized a 25-year old software suite with 550,000+ lines of hand-written Java code as a sole developer using Claude Code and multi-LLM spec-driven development. Java 6 to 11, Eclipse RCP 3.6 to 4.19, Ant to Maven, and a new interactive React reporting module that runs in the cloud and embedded in the desktop app. 1,294 commits, 791 automated tests.
How I built two agentic AI systems for a 25-year-old load testing application: one that configures test cases through natural language conversation, and another that agentically investigates load test data and generates professional performance analysis reports.
How I built a natural-language reporting system for personal finance using AI, reverse-engineering the Copilot Money API, and creating a Chrome plugin with MCP server capabilities.