A Designer’s “Impossible Mission”: Reviving a Classic RTS with AI
Ammaar Reshi, Head of Product and Design for Google DeepMind AI Studio, has completed a remarkable personal project: natively porting the classic 2003 PC game ‘Command & Conquer: Generals – Zero Hour’ to the iOS platform. Notably, Reshi’s professional background is in product design, he majored in poetry in college, and has no C++ programming experience. He led the entire porting process by giving instructions to an AI model.
His core tool for this project was the Claude Fable 5 model from Anthropic, a competitor to Google. Reshi stated that this demonstrates that professionals can maintain a passion for the entire AI field, respect competitors’ products, while being dedicated to building their own best solutions. This isn’t his first time using AI for creation; in 2022, he used ChatGPT and Midjourney to create a children’s book titled ‘Alice and Sparkle’, which sparked widespread discussion.
Bridging a 20-Year Tech Gap: From DirectX 8 to Metal
The technical challenges of this port were immense. The game’s original engine contains about 1.6 million lines of C++ code and relies on Microsoft’s DirectX 8 graphics API. To run on a modern iPhone, its rendering commands had to be translated to Apple’s Metal API. With AI assistance, Reshi built a complex five-layer translation chain:
- DirectX 8 (Original API)
- DXVK (Translates DirectX calls to Vulkan)
- Vulkan (Cross-platform graphics API)
- MoltenVK (Translates Vulkan calls to Metal)
- Metal (Apple’s native graphics API)
Each layer of translation risked introducing errors, performance degradation, and compatibility issues. Additionally, the project required replacing several Windows-specific components, such as the Miles Sound System for audio, Bink for video decoding, and GDI for font rendering, with cross-platform solutions. After switching to Claude Fable 5, the first successful build took only about 40 minutes.
AI as Not Just a Coder, but a Debugging Expert
Successfully rendering the game was just the first step. To adapt to a mobile device, Reshi used AI to re-engineer a complete set of native touch controls to replace the original keyboard and mouse setup. This new system supports gestures like tapping to select units, dragging to box-select, long-pressing to cancel, two-finger scrolling, and pinch-to-zoom, resolving issues with accidental inputs caused by imperfect gesture recognition in early versions.
During the debugging process, the AI also helped identify and fix several deep-seated issues:
- Hardware Recognition Error: The game engine couldn’t recognize the modern “Apple A19 Pro” chip, incorrectly setting the graphics quality to the lowest setting. The solution was to bypass the auto-detection logic by force-writing a high-quality configuration file on the first launch.
- A 20-Year-Old Hidden Rendering Bug: An error in the texture format fallback logic, which never manifested on PC, caused the game’s minimap fog of war layer to render as a solid black block on iOS due to differences in format support by MoltenVK. The AI helped locate and fix this legacy bug.
The Rebirth of a Gaming Legacy: Synergy of Open Source and AI
This project was made possible by a significant move from publisher Electronic Arts (EA) in February 2025. Driven by Game Preservation Project Lead Luke Feenan, EA open-sourced the source code for several ‘Command & Conquer’ titles, including ‘Generals’ and its expansion pack, under the GPL v3 license. This open policy paved the way for community modifications and platform ports. In contrast, EA’s loss of the ‘Red Alert 2’ source code means it cannot be revived in the same way.
Reshi’s project showcases a new model for game preservation and redevelopment: code is open-sourced by the publisher, combined with the passion of community developers, and empowered by AI to dramatically lower the technical barrier and development cycle.
Achievements, Limitations, and Future Outlook
Currently, the project is open-sourced on GitHub and can natively run core modes like Campaign, Skirmish, and the Generals’ Challenge on ARM64-based iOS devices. However, it faces some practical limitations:
- Complex Installation: The project is not available on the App Store. Users must configure Xcode and the Vulkan SDK themselves, compile from source, and sideload the app.
- Memory Consumption: Due to the overhead from the multiple graphics API translation layers, the game’s memory usage can exceed 3GB on an iPad during long sessions, potentially leading to it being forcibly closed by the iOS system.
- Lack of Online Multiplayer: Because it’s difficult to guarantee identical results for floating-point calculations across different CPU architectures, the lock-step synchronization mechanism the original engine relied on cannot be implemented, making online multiplayer unavailable.
Despite these limitations, the fact that a non-programmer completed such a complex software port in just two days (using up an entire Claude Fable 5 Max subscription) clearly signals the immense potential of AI in the future of software engineering. It proves that with AI assistance, individual developers or small teams will be able to tackle complex projects that once required large, specialized teams.
