The technological barrier that protected dusty legacy code from modernization for decades has finally collapsed. What used to require months of manual labor from an entire porting team has been reduced to a few hours of lazy supervision over an AI agent. Ammaar Reshi, Head of Design at Google AI Studio, recently demonstrated this shift in power: he successfully ran the 2003 classic *Command & Conquer: Generals Zero Hour* natively on iPhone and iPad. No clunky emulators involved—just the raw enthusiasm of Claude Code and Fable 5.

The project—migrating from x86 and DirectX 8 to the ARM64 architecture and Metal graphics API—is more than just "running an old game." It is a deep dissection of system logic. Anthropic’s agent produced the first build in just 40 minutes, and over the next couple of days, it polished it into a fully functional state with native touch controls, working campaigns, and skirmish mode. The economics of the process look terrifying for outsourcing firms: instead of bloated payrolls, the costs were limited to a single exhausted Claude Max subscription limit.

"This case proves that architectural barriers have turned into thin partitions that modern agents simply kick down."

Ironically, a Google DeepMind employee used a direct competitor's tool to achieve these results. This underscores a painful fact for Google: in the autonomous coding race, Anthropic is currently setting the pace while others are still trying to figure out how to embed chatbots into IDEs. Reshi’s case proves that agents have outgrown generating simple scripts and are now capable of refactoring complex graphical pipelines.

Main takeaways from the case:

Speed: The first working build of a complex project was obtained in less than an hour. Cost: Migration expenses dropped from hundreds of thousands of dollars to the price of a monthly AI subscription. Quality: The agent successfully rewrote the graphics API from obsolete DirectX to modern Metal. Autonomy: The AI independently implemented touch control support that did not exist in the original.

While iPad memory leaks still require manual intervention, the architectural heavy lifting is now handled by neural networks in one sitting. For R&D heads and CTOs, this precedent changes the rules of the game for technical debt. Bulky, twenty-year-old code is no longer a lifetime liability requiring million-dollar migration investments. Now, it is simply a weekend task for a single engineer armed with the right agent. If your software is still stuck in the Windows XP era, you have officially run out of excuses.

AI AgentsDigital TransformationCost ReductionAnthropic