Human direction. Inspectable work.
Practical AI-assisted development, from a clear product goal to software that can be reviewed and tested.
3 case studies
Case study · Working prototype
How I directed AI-assisted development of Unproven’s Inventory Core, from bounded requirements to independent review and human acceptance.
C# / Unity / ChatGPT / Codex
Read the case studyCase study · Working prototype
A placeholder-first Unity browser RPG. I directed the mechanics and playtested AI-assisted builds; player feedback exposed unclear objectives and invisible attacks that passing checks had missed.
Unity / C# / WebGL / Codex
Read the case studyCase study · Working prototype
How I directed an Astro website through design decisions, bounded implementation, artwork optimization and browser review, while keeping publication separate from a working local draft.
Astro / TypeScript / CSS / Node.js / Codex
Read the case studyHow I work
I set the experience, priorities and boundaries. I work with AI on requirements and architecture, then give coding agents bounded implementation tasks. Review, test results and hands-on use inform what I accept.
The amount of process should fit the risk. Persistent state and authority boundaries need careful specifications; a disposable visual experiment can use a lighter brief and a direct review.
These studies describe the actual work and its limits. Passing checks are evidence about their scope, rather than a guarantee that a player will understand the result. The website-build study applies the same process to the site you are reading.