Skip to content

AI Apps as Operating Systems

January 9, 2025 at 03:22 PM

Note: This is not a blog, it's a semi-private digital garden with mostly first drafts that are often co-written with an LLM. Unless I shared this link with you directly, you might be missing important context or reading an outdated perspective.


Mobile AI apps like Claude and ChatGPT are evolving from chat apps into something that more closely resembles operating systems. One simple but powerful way to conceptualize the space is that computers as a whole and apps more specifically are just giant functions that take in inputs and return outputs.

Right now they’re very good at taking in fuzzy human input and making sense of it (especially in voice and text), but still need other apps to really nail output.

Other relevant convergent themes are Artifact Centric Computing and Audio-UIs. Also, tangentially related are, Slow Computing and Computation without Computers.

The Current Landscape

The major players each have distinct approaches:

The Platform Strategy

A key tension in this space is between being open vs. insular. While these apps currently provide 80-90% of needed functionality within their interfaces, their approach to external integrations varies:

The Data Retention Strategy

A critical aspect of these AI operating systems is their approach to user data retention. As users generate more complex knowledge structures through their interactions:

This “data moat” strategy mirrors traditional operating systems, where the difficulty of migrating years of accumulated data and workflows becomes a powerful retention mechanism.

The Developer Opportunity

The key to success in this space may lie in creating a low-code environment where developers can:

Raw

I want to add a note to the blog post I have about AI apps as operating systems. The retention play, of course, here is in data and more and more complex data structures For example, projects where users can not only store a bunch of their threads but then start asking questions about the knowledge they've generated the thread already. At this point, we're getting to the level where only very complicated users will be able to pull data out.