A Google DeepMind developer just rebuilt a 20-year-old strategy game for the iPhone using Anthropic’s Claude Code, and it took an afternoon. According to The Decoder, Ammaar Reshi, Lead Product and Design for Google AI Studio, ported “Command & Conquer: Generals Zero Hour” to iPhone and iPad with help from Claude Code and Fable 5. The 2003 real-time strategy game now runs natively on ARM64. No emulator, no cloud streaming, just the actual game running on Apple silicon.
What stands out here is the speed. The first working build took about 40 minutes, Reshi says, followed by “a few hours” of debugging. Campaign, skirmish, and the “Generals Challenge” mode all work, remapped to touch controls. Over two days he burned through his entire Claude Max quota, which tells you how much heavy lifting the AI was doing.
What actually got done
- Graphics translation: The game was built on DirectX 8, which Apple hardware doesn’t speak. The pipeline converts DirectX 8 calls to Apple’s Metal API through several intermediate steps.
- Architecture jump: The code now runs natively on ARM64, the chip family in every modern iPhone and iPad.
- Input redesign: Mouse-and-keyboard controls got reworked for a touchscreen.
- Documentation: An engineering log captures every bug and every fix along the way.
Reshi published the full source code on GitHub as open source. The game assets aren’t included, so you need your own copy to play. It’s on Steam for around $5. Fair warning from The Decoder’s report: on iPads, long sessions can crash the game because of high memory usage. This is a working proof of concept, not a polished App Store release.
Why this matters
Game porting has always been slow, specialized work. Studios budget months and dedicated engineers to move a title from one platform to another, especially when the graphics layer has to be rewritten from scratch. One person doing a rough but functional port of a DirectX 8 game to native Metal in a couple of days reframes what that work costs.
The interesting part isn’t that AI wrote some code. It’s the category of problem. Translating a graphics API, targeting a different CPU architecture, and untangling 20-year-old C++ is exactly the kind of deep, unglamorous systems work people assumed would stay human for a long time. Reshi’s project shows a coding agent can carry a lot of it, provided someone experienced is steering and debugging.
There’s a competitive wrinkle worth noting. Reshi works at Google, yet he reached for Anthropic’s tool. Asked about that, he said “you can love the AI space and respect the competition while still being fully focused on building the best answer. It’s a long game.” That’s a quiet endorsement of Claude Code’s coding ability from someone with every reason to use his own employer’s product.
What to take from it
A few practical signals for anyone building with these tools:
- Agents handle systems work now, not just scripts. API translation and cross-architecture porting are in reach, not just web apps and boilerplate.
- The human is still the bottleneck and the safeguard. The AI produced a first build fast, but hours of debugging came after. Judgment, testing, and knowing what “working” means still fall to the person.
- Quota is the new constraint. Burning a full Claude Max allotment in two days is a real cost signal. Ambitious agent projects consume tokens fast, and that shapes what’s economical.
- Open source lowers the bar for the next attempt. With the code public and the engineering log documented, the next person porting an old game starts miles ahead.
Expect more of this. If one developer can revive a 2003 title over a weekend, a wave of hobbyist ports, preservation projects, and platform experiments is coming. The bigger question is what happens when studios point the same tooling at their own back catalogs. Full details, including the GitHub repo and the engineering log, are available at the original report from The Decoder.