The AI Workflow That Lets Me Code My Game (As a Non-Programmer)

👉 Wishlist my game on Steam: https://store.steampowered.com/app/4003040/Dino_Card_Hunt/ In this devlog, I share a major update to my Unity workflow for building my roguelite pet battler deckbuilder. I'm moving beyond simple prompts and implementing a "Sub-Agent" system based on Anthropic's own best practices. As a game artist and designer, not a programmer, this new approach allows me to architect complex features with more accuracy and control. I’ll take you through the entire process of adding a brand-new "Silence" card to the game, from the initial design document to the final, working implementation, all orchestrated through Claude Code. We'll see if this powerful, token-heavy method is a glimpse into the future of indie game development. 💡 In this devlog, you'll see: My new multi-phase workflow: Explore, Plan, Implement, and Validate. How I use AI "Sub-Agents" for parallel research and code exploration. The importance of creating a simple design document to guide the AI. Why I believe the developer must always be the architect, not the AI. A live demonstration of creating a new "Silence" card and effect. Managing the high token cost and potential limits of this advanced workflow. My process for debugging when the AI's first attempt isn't perfect. ⏱ Timestamps: 00:00 – My Updated Workflow 00:21 – The Importance of Refactoring (SSOT & DRY) 00:57 – Current Gameplay Demo 03:02 – Creating the "Silence" Effect Design Document 04:57 – Building a Prompt Template from Anthropic's Docs 07:02 – Explaining the "Sub-Agent" Parallel Workflow 12:13 – Live Implementation: Running the Sub-Agents 15:56 – Adding the New "Silence" Card and Assets in Unity 17:23 – First Test: Success! The Silence Card Works 19:30 – Identifying and Debugging Lingering Issues 22:50 – Rerunning the Prompt to Fix the Bugs 24:15 – Final Demo: The Polished Mechanic Works 25:29 – Final Thoughts on My Non-Programmer Workflow