Note: This post is a work in progress. More coming.
We've been applying AI from the top down. We actually need to be thinking about how we apply it from the bottom up.
Essentially, imagine an operating system that is blank when you buy the computer. And all you have is an agent that has a somewhat lightweight, maybe a medium-weight model. Or both. They can take actions and do a bit of thinking and do a bit of creating. And that agent essentially writes the OS itself.
We've seen Claude writing Linux kernels. We've seen it writing C compilers. Why can't it just do this live, eventually? For you. Write exactly what you want into the OS-level architecture. When you need it.
Once we compress the inference needed to do things like this at an OS level, it changes everything. Maybe it even creates a programming language specific to each person to more optimally do the things that person asks it to do. Solidifying those neurons. And that's why we need an AI operating system.
We are so far past the time of user interfaces. Now there's a user experience. Now it's all middlemanned by AI. Because it can provide the exact experience that anyone wants.
So it's up to everyone to provide the capabilities of whatever service they offer to the agents. And then you can take those capabilities and have your agent orchestrate them however you need. Stitch them together, wire them up, just by telling your agent what you want to accomplish. The same can be true for an operating system. And honestly, nothing speaks truer to the term "operating system" than this. Because it's how you operate your system.
Now if you think about this in terms of bottom up, what it actually is is this. You open up a fresh computer. And say you are an author and you like writing. And you like writing in a certain setting. You can tell your system now to essentially give you the experience of writing in that type of setting. You want to be focused on the page. You want to be able to just talk through your thoughts. You want to be able to have that written and summarized on paper. And that allows you to come in and then restructure things the way you want them to be by just telling the AI how to write it.
As you use your devices, your devices are essentially growing neurons that are solidifying as the system realizes what you really like. And what you don't. And it can forget just as fast as it learns. And that's extremely important.
So in the end, you build this system around what you love doing, and then it's malleable. And it learns what you like more, what you don't like. And it turns into this extension of you. Your abilities, your process. And it just enhances everything you do daily. The services you provide, the way you think.
You can then operate at the speed of thought.
It can predict things that you would want in certain situations. It can have defaults ready. DHH talks about great defaults on the Lex Fridman podcast. I believe a lot in that. And now you can have AI architecting those defaults for you. As it gets to know you. As those neurons solidify.
Gone is the era of user experience. Because you now craft your own user experience down to the operating system level. You don't need to know how to write your own operating system. Because it can literally write the code for you.
That's why Anthropic and OpenAI went for writing code first. Because that's what architects our technological world. And we need to offer that to everyone in the world. That's what's going on.
As soon as everyone in the world can architect their own experience around the capabilities offered by the services around them, we'll have a fully agented world. A world of technological abundance.
Ideas instead of apps. Processes instead of applications that you download.
No more "what kind of to-do app are you using?" It's "what kind of process do you use when you keep track of tasks you want to do?" Oh, that's a cool process. Well, you know, I might adopt some portion of that into what I do. And my ecosystem will update.
And actually, there's no more ecosystem. It's just your ecosystem. What you build, what you use on a daily basis, what you want to see at every given moment of what you're working on. What your workspace looks like at any moment that you're feeling a certain way, down to exactly what you want.
You're wearing an Oura ring that tracks your heart rate. Well, guess what? What if you want to essentially have your computer help you calm down at certain times? Maybe you have a medical disability of some kind, and this would allow you to attend to that easily in your process. Because it learns that about you. It learns what you like, what you need, what makes you operate at your max capacity.
Ever since AI came out, we've been gathering data and information on how it's used to understand what people would first use AI for. And that's really just thinking about what should we tackle first once we get to the point that AI can do so many things. That phase generated the revenue, the money, and the capital in order to have what Cursor and Anthropic and OpenAI have now. These autonomous agents just coding and coding and coding. And sure, it takes weeks now. Later, it's going to take seconds.
When it takes seconds, it's going to be amazing.
Imagine your laptop with your small to medium-sized models that can take action or do a bit of thinking for small tasks. And then it realizes it needs more processing power. So then it takes after the Apple Intelligence architecture and infrastructure where it offloads deeper thinking into the cloud. Kind of how Anthropic has been structured. Haiku to Sonnet to Opus. Haiku and Sonnet may be able to have their own localized onboard models.
My big prediction of the next five years, maybe even sooner, is that Anthropic is going to buy Apple.
Because Apple has created this hardware manufacturing line of the best hardware in personal computing. So now it's all about taking that hardware and applying the right software to it. And that's going to be AI-first, AI-native software. Which Anthropic is building.
So if Anthropic buys that hardware pipeline and puts Haiku and Sonnet on device, Opus in the cloud, now everyone who picks up that machine can build their own ecosystem of apps and tools around exactly what they want, when they need it.
That's the type of computing that is coming.