The Architect: AI-Driven Planning for New Game Features
In this article, I’ll show you how to use an AI assistant as a powerful architectural consultant to plan complex new game features, analyze options, and generate detailed implementation roadmaps.
In the first article of this series, I covered how to prepare a project for AI collaboration by creating a CONTEXT.md file. Once you have that foundation, you can move beyond simple code generation and use the AI for one of its most powerful applications: architectural planning.
Before writing a single line of code for a new feature, you can leverage an AI assistant to explore technical options, weigh trade-offs, and create a comprehensive implementation plan. This dramatically reduces risk and ensures you’re building on a solid foundation.
The Methodology: From Idea to Workplan
I use a four-step methodology to guide the AI from a high-level feature idea to a detailed, actionable workplan:
- Define the Problem: Clearly and specifically state the feature you want to build and the technical constraints.
- Propose Solutions: Ask the AI to analyze the problem and propose multiple implementation strategies.
- Build a Workplan: Once you’ve chosen a solution, ask the AI to create a granular, phase-by-phase implementation plan.
- Review and Validate: Critically review the AI’s plan, ask clarifying questions, and approve the final strategy before starting to code.
Let’s explore this with two real-world examples from the MonoRpg project.
Case Study 1: Choosing a New Map Editor
The first major architectural decision was migrating from the old Tiled map editor. Instead of just picking a new one, I used the AI to perform a comparative analysis.
1. Define the Problem My goal was to find a modern map editor that offered a better designer experience, type-safety for entity properties, and a more unified workflow.
2. Propose Solutions I prompted the AI with the following:
“Compare 4 approaches for map editing in my MonoRpg project: a web-based editor, a custom standalone editor, an in-game editor, and adopting the LDtk editor. Analyze each based on development cost, maintenance, quality, and integration effort.”
The AI generated a detailed comparison matrix that made the trade-offs crystal clear. It correctly identified that building a custom editor was a huge undertaking, while adopting LDtk offered professional-grade features for zero tool development cost.
3. & 4. Build Workplan and Validate Based on the AI’s analysis, I made the decision to adopt LDtk. The next step, which I’ll cover in a future article, was to have the AI generate the detailed migration workplan for this chosen solution.
Case Study 2: Designing a TCG Card System
The second example was even more complex: designing a complete Trading Card Game (TCG) combat system to be integrated into the RPG.
1. Define the Problem The prompt was:
“Design a TCG card system for MonoRpg. We need card graphics, mechanics, an editor workflow, ECS integration, and support for both solo and future multiplayer modes. Propose an approach that minimizes tool development cost.”
2. Propose Solutions The AI’s response was a comprehensive, 100+ page design document proposing a Hybrid Data-Driven Approach. It recommended:
- Graphics: Use Figma/Photoshop for high-quality card templates.
- Data: Define card mechanics and stats in simple JSON files.
- Editor: Defer building a custom editor, as the JSON files could be edited by hand.
This approach perfectly aligned with my goal of minimizing custom tool development.
3. Build a Workplan With the architecture chosen, I prompted:
“Create a detailed workplan for implementing the Hybrid Data-Driven TCG system. Break it down into phases with specific tasks and files to modify.”
The AI generated a 14-phase plan, covering everything from the ECS component design to the game mechanics and persistence layers. It even designed the system with a ICardPersistenceService interface, ensuring that adding multiplayer later (by creating a RemoteApiCardPersistence implementation) would not require a major refactor.
4. Review and Validate I reviewed the entire plan, validated that it fit my existing architecture, and approved it. The result was a clear, actionable roadmap for a massive new feature, created in a fraction of the time it would have taken to plan manually.
Conclusion: The AI as Your Technical Co-founder
Using an AI assistant for architectural planning is like having a tireless technical co-founder to bounce ideas off of. It can explore avenues you hadn’t considered, perform detailed trade-off analysis, and structure your thoughts into a concrete plan.
This process doesn’t replace the need for human expertise—I still made the final decisions—but it augments it dramatically. By investing a few hours in planning with an AI, I saved weeks of potential rework and started development with a high degree of confidence.
In the next article, I’ll show you how to execute on these workplans, using the AI as a pair programmer to tackle large-scale refactoring and feature implementation.