---
title: "Desktop, Touch, Browser, Now AI? The Next OS in Computing"
description: "Explore how AI is reshaping human-computer interaction, from command lines to modern SaaS. Key insights on the next computing paradigm for startups and tech leaders."
categories: ["AI"]
keywords: ["AI interaction","human-computer interaction","SaaS computing","Tomasz Tunguz","desktop operating systems","AI as an OS","command line interface","Llama3.3","venture capital","AI-driven productivity"]
ai_summary: "Discover how AI is transforming human-computer interaction and shaping the future of operating systems."
date: 2024-12-16
lastmod: 2026-07-31
canonical_url: https://www.tomtunguz.com/the-ai-os/
author: "Tomasz Tunguz"
---

Remember the first time you touched a computer screen instead of typing commands?

We've lived through distinct epochs of human-computer interaction: the cryptic beauty of command lines, the intuitive dance of graphical interfaces, & the ubiquity of browser-based computing in the SaaS era. 

It's different now. When I manage spreadsheets, I don't want to manipulate formulas anymore. Instead, I want to instruct the computer as I would explain to a colleague : "run the correlations on these variables to see if anything meaningful pops out, then plot it, & add it to the deck."

When I parse research reports, I don't want to read them line by line ; I want to ask for summaries & surprising conclusions. 

When I program computers, I don't want to work at the level of lines & functions & arguments, but the style of a webpage like this blog's theme.

When I read an email, I want to command an AI to follow up on this in a week, or research the new prompt engineering technique in this newsletter & send it to my Kindle so I can read it that night, or compare the revenue multiples on health care records companies to horizontal CRM.

So, I've taken to keeping an AI open on a separate monitor. That's step one. 

But I'm starting to run models on my laptop so I can fire all kinds of questions at its feet & watch it dance. Can it draft an email, or critique a blog post well? 

Running [Llama3.3](https://ollama.com/library/llama3.3) on my computer is the nearly same as asking any of the major alternatives - but more private & a bit slower - which for the moment works for me. It buys me time to read the AI's output and reflect on this new working relationship.

Talking to the computer using the [transcription program](https://tomtunguz.com/ai-use-in-one-afternoon/) I built (that I'll soon share), reinforces the sense of collaboration.

I can see a world where I'm not flipping between applications. Instead, I'm telling the AI to send an email or research multiples & that's my interface - that's my OS. Not the command line, not the desktop, not the browser. 

The AI choreographs all that work underneath.