Microsoft reimagines app creation with Copilot

Microsoft's CEO Satya Nadella introduced a new concept at the Windows and Surface event on October 7, 2026, allowing users to request their assistants to generate applications instead of downloading them from a store. This new approach uses Copilot, powered by advanced models such as MAI-Code-1.1 Flash, which is optimized to run on personal computers.

MAI-Code-1.1 Flash model details

The MAI-Code-1.1 Flash model has 137 billion parameters and a 256,000 token context window, allowing it to generate applications, dashboards, or workflows from natural language. The model has been reduced in size by 80% to 3 bits, making it more efficient for local execution on a PC.

Local execution and cloud integration

Although the model runs locally on the device, it does not operate entirely offline. Tasks are processed via GitHub HydraFusion, which decides whether to execute them on the device or in the cloud. The local version of the model weighs 53 GB and requires up to 75.5 GB of memory.

Hardware requirements and cost

The tested Surface Laptop Ultra (RTX Spark) requires a minimum of 64 GB of memory, with only the 128 GB configuration meeting this requirement. This model costs €6,879 and is not yet available for purchase. These high specifications may limit widespread adoption.

Economic and technical implications

This innovation could shift how enterprises view software subscriptions if AI can quickly replicate specialized tools. However, generating functional applications does not solve issues related to maintenance, security, or long-term compatibility. This may affect long-term software reliability and user trust.

Strategic impact on app stores

If Copilot becomes the primary interface, Microsoft might shift user relationships to its assistant, which could change the dynamics of app stores and distribution models. This could potentially reduce the role of traditional app stores in software distribution.

Apple's response to similar challenges

Apple has faced similar issues with apps like Replit and Vibecode, which allow users to write and run code. Apple has restricted such apps due to its App Review Guidelines, but it has proposed a compromise by allowing previews in external browsers. Apple also integrates AI tools into Xcode for code generation, though these may produce web apps that bypass App Store regulations.

Performance comparison between Microsoft and Apple

Microsoft claims its RTX Spark PCs are faster in generating the first token, images, and videos compared to Apple's MacBook Pro 16-inch M5 Pro. However, these tests were conducted on pre-series machines with specific models, which may not reflect real-world performance. Enterprise users remain cautious about the reliability and data security of AI-generated tools.

Microsoft's safeguards for AI-generated tools

To address concerns about data leaks and reliability, Microsoft uses execution containers (MXC) and administrative policies to restrict agent access. These measures ensure regular updates and accountability, helping to maintain user trust and system security.