AI

I’ve been delegating more coding than ever to AI in the last few months. Over time, I noticed some common patterns that work and common pitfalls. I’ve distilled what works for me when using AI for most of the heavy lifting. This isn’t about prompt engineering tricks—it’s about fundamentally rethinking your development workflow. If you’re an experienced engineer looking to delegate more coding to AI while maintaining quality and control, this guide will help you assess where you are and where you could be. ...

I heard about Ralph Wiggum as a Software Engineer a couple of months ago from Geoffrey Huntley’s article. The idea was intriguing and I wanted to build something real with it. At the same time, I’ve had a feature on the backlog of Sharp Cooking for about a year now. I knew it wasn’t an easy one and it was never a critical feature either. What a great opportunity! I highly recommend reading the original article as well as the snarktank/ralph implementation that was created since. The idea is simple yet powerful: create a detailed spec, set up the AI agent, let it run and learn in an autonomous loop until it completes the spec implementation. This post documents my experiment that became a production feature. ...

SpecKit is an open-source toolkit from GitHub that transforms how we build software by putting specifications first. Unlike traditional development where we jump straight into coding, SpecKit guides you through a structured process: define what you want, clarify requirements, create a technical plan, break it into tasks, and finally implement. While it provides a CLI for setup, it’s primarily a set of templates and agents that work within your existing AI coding assistant like GitHub Copilot. ...