The Future of Operating Systems in the Age of AI

The impact of generative AI on computing is shifting from the first-order effect of faster coding to a second-order effect: the dissolution of the boundary between the programmer and the user. As natural language becomes a viable programming language, the traditional model of distributing prefabricated, fixed-function applications created by professional developers is becoming obsolete.

The Shift from Prefabricated Apps to Conjured Software

Traditional software distribution has evolved from physical media (boxed software, CD-ROMs) to the internet. The next evolution is a move away from fixed-function applications entirely. In a world where users can "conjure" software using English, the audience for any given application may shrink to a single person.

This shift enables the creation of highly idiosyncratic tools to solve trivial, personal problems that would never have been commercially viable for a professional developer to build. Examples include:

Hyper-local weather forecasting: Creating a tool specifically for a neighborhood (e.g., Roscoe Village in Chicago) to account for lake effects that regional stations miss.
Context-aware navigation: A tool that provides specific routing advice for a specific commute (e.g., the Eisenhower Expressway) without the overhead of a full map interface.
Personalized scheduling: A tool that monitors meetings across various platforms to notify the user of immediate obligations.

Redefining the Operating System

Historically, the core purpose of a modern operating system (OS) has been to partition different applications from one another and control their communication. This security model is based on the assumption that software is imported from strangers.

However, this model becomes less relevant when the majority of the software on a device is created by the user themselves or by people they trust. When applications are malleable and subject to constant change on the whim of the creator, the rigid "fiefdoms" of current OS process isolation may no longer be the optimal way to manage hardware resources.

The Hardware Gap: The 2026 Smartphone

Most smartphones available in 2026 were designed in 2023, a time when computers were still viewed as devices for running deterministic, fixed-function applications. The current smartphone architecture is essentially a miniaturized version of the 1970s design—built to run prefab apps efficiently but not designed for a world where the device itself builds the apps it runs.

Technical Counterpoints and Industry Challenges

While the vision of a "malleable OS" is compelling, several technical and systemic hurdles remain:

1. The Role of the OS as a Shared Database

Some argue that the OS provides value far beyond process isolation. Modern OSes (like iOS and macOS) act as shared system-provided databases (e.g., HealthKit, Contacts, Calendar). These services provide a secure, unified fabric that allows disparate apps—even those generated by AI—to share data consistently.

2. Security and Trust in Third-Party Services

Even in a user-generated app ecosystem, the need for professional software remains for high-stakes tasks. Banking, medical records, and payment systems require signed binaries and OS-level guarantees of isolation to ensure security and trust. A totally malleable OS may struggle to provide the necessary guarantees for these commercial services.

3. The "Agent" vs. "App" Debate

There is a theoretical divide between creating a personalized app and using a single AI agent. Some suggest that the future is not thousands of personalized apps, but a single AI assistant that completes tasks directly without the need to manifest a discrete application interface.

4. The Bootstrap Problem

Historically, new operating systems have struggled with the "bootstrap problem": the need for a critical mass of users to attract developers, and a critical mass of apps to attract users. While AI-generated apps might bypass the need for professional developers, the system would still need to integrate with the APIs of major third-party services to be viable.

Historical Precedents

This movement toward a programmable, reflective system echoes early computing experiments, such as the Xerox Alto and the MIT Lisp machines. These systems often booted into a single address space where the OS and applications co-existed, prioritizing flexibility and productivity over the strict security boundaries that define modern computing.