Ammaar Reshi, a product and design lead at Google, shared a demonstration showing The Elder Scrolls V: Skyrim, Batman: Arkham City, Hades, and Age of Empires II running on an iPad mini. Reshi says these are the original Windows versions, with touch controls added, and that the games run locally rather than being streamed from a PC.
Astra Was Given the Tedious Engineering Work
The interesting part is not simply that an iPad can run demanding games. It is how much of the normally labor-intensive compatibility work can now be delegated to an AI agent.
Reshi said he normally supervised Astra’s work and provided feedback on game performance. For one experiment, however, he went to bed after instructing the model to get Batman: Arkham City running on the iPad by the time he returned. The resulting demonstration suggests Astra was able to work through installation, configuration, debugging and compatibility problems with limited human intervention.
That matters because running a Windows game on iPadOS is considerably more complicated than simply installing an application. A compatibility stack may need to translate processor instructions, reproduce Windows APIs, translate graphics calls and adapt keyboard or mouse input for a touchscreen environment.
Four Games Test More Than Raw Performance
The selection of games is also notable because they represent different workloads. Skyrim and Batman: Arkham City are older PC titles with demanding 3D graphics. Hades presents a different modern game-engine workload, while Age of Empires II creates a different challenge because its traditional mouse-driven interface needs to work with touch input.
That makes the demonstration more significant than showing a single carefully optimized game. It suggests the underlying approach could potentially support a wider range of Windows software, although the current evidence is still only a demonstration rather than proof of broad compatibility.
The Real Breakthrough Is Software Engineering
OpenAI introduced GPT-6 Astra as a model focused heavily on computer use and software engineering, including the ability to interact with software environments rather than merely generate code. The iPad experiment provides an unusually concrete example of what that capability can mean outside conventional application development.
Our Take: The important development may not be Skyrim running on an iPad. It is the possibility that AI agents can make previously uneconomical compatibility projects practical. Bringing an old Windows application to another platform traditionally required specialist engineers, extensive debugging and enough commercial demand to justify the work. An agent that can repeatedly install, test, inspect failures and modify the environment changes that equation.
There are still major questions. The demonstration does not establish long-session stability, thermal performance, battery consumption, compatibility across iPad generations, save-game reliability or how many other Windows titles can use the same environment. The exact implementation is also not yet publicly reproducible.
Reshi says a repository and setup guides are coming. If those materials demonstrate a repeatable process rather than a one-off configuration, the experiment could become much more consequential. It would show that AI agents are moving beyond writing software and increasingly becoming capable of adapting existing software to environments its original developers never targeted.
Follow us on Bluesky, LinkedIn, X, and Telegram to Get Instant Updates
