AI-Assisted Engineering

Most of the software I build now, an AI agent does the typing and I do the specifying and the review. If you build software right now, you are probably somewhere inside the same shift, whether or not you chose it.

I started writing about it in mid-2025, after six weeks working almost entirely in agent mode. The speed was real, but it was the smaller change. The bigger one was in how I thought: how I scoped work, and how I decided what to do myself and what to hand off. That reached all the way into the shape of the development process.

The essays here are field notes from that experiment, which is still running. They cover the workflows that held up, the failure modes I did not see coming, what working at machine speed did to my thinking, and what happened when I used the same patterns on infrastructure, a content pipeline and business software with production requirements.

AI-assisted development asks for more engineering judgment, not less. The models can write code. They cannot decide what to build, catch the architectural mistake that turns into a rewrite six months later, or know when good enough is good enough. That judgment is still mine, and the point of the tools is to leave me more time and attention for it.

Start with the current work just below. The 2025 series that started it is collected in one overview further down, if you want the whole arc. Every piece stands on its own.