It is a written adaptation of a conference talk that explores how to make game projects more 'legible' to AI agents by treating the project itself as a harness for agentic iteration. It covers LLM prompt fundamentals, why non-code assets make projects opaque to LLMs, and techniques for building faster agentic development loops.
Most valuable when integrating LLM agents into complex, asset-heavy, or non-standard web/game projects, or when agentic coding tools repeatedly fail on your codebase and you need to understand and fix the underlying causes.