I am a professional software engineer by trade. Like in most companies, I am required to leverage AI coding tools to do my work. I have been doing so for over a year, and in the last few months I no longer write or review any code by hand.

For most of my game projects I write the code myself "for fun" as a fading opportunity to practice the mechanics of my craft. But on this project I decided to build it the same way I build software at work. I want to share my approach to show you what it is really like.

The full write up is at https://blog.jeffschomay.com/how-i-m-using-ai-for-game-dev-in-2026 -- please read that for the full picture and links to the exact agentic harness and document files that I used.

TL;DR

Flow: design doc sketch > enter claude > flesh out full design doc > break into milestones > iterate over milestone implementation > manual playtesting > agentic playtesting

What adding an extra milestone looks like:

• 1hr outlining idea to turn into milestone
• 2hrs editing milestone
• 30m claude implement autonomous
• 5hr playtesting and tweaking with claude interactively (improved with new agentic playtesting approach)