Recently, Andrej Karpathy gave a talk at YC startup school, called “Software is Changing (Again).” In it, he talks about how AI, specifically LLMs, have created an entirely new breed of software, with reasoning at the core of it, and human language as the primary way that we program that software.
As he explores this idea, he compares LLMs to other things, like utilities and fabrication plants, and lands on what he thinks makes the most sense: an operating system.
Earlier today, as I was using ChatGPT, it hit me what this means, and why it’s a fantastic comparison. Take a look at the ChatGPT UI. It’s a new type of operating system, with the same metaphors we have today:
- Your past chats are up in the terminal (for the non-nerds, that’s how you look back through past commands you gave the computer)
- The GPTs, and specialized things like Codex and Cora are apps
- The Projects are your filesystem
- And the main chat interface is the primary UI into your information, and the world
ChatGPT isn’t an app, it’s an operating system, and the pattern that was developed by OpenAI has now become pervasive. This is exactly how Gemini, Claude and X work. It’s how all AI codegen apps work. This is the new pattern for computing in the future.
So what does this mean?
Well, it probably means a lot of things, but the most important thing to recognize is that this is a major shift that will likely end up redefining most of our interactions with people, companies and information through the internet. Just like desktop software nearly ubiquitously moved onto the cloud and became accessed through the browser, so will today’s SaaS and cloud apps move into this new space, with an entirely new form factor and user experience.
This is not new. We’ve seen this transition happen a lot actually, from command lines in the 80s, to desktops in the 90s, to the web in ‘00 and (partially) to apps in the ‘10s. This is another modality shift, and while it won’t completely wipe out the existing modalities of the web, desktop and mobile, it will likely eat into a sizeable share of that audience, if not for any other reason, just because of how naturally accessible it is. You just talk to it, and it works.
For everyone in the tech industry, this should be a wake-up call. You’re getting a chance to be a first-mover in a wave you can see coming. This is like being able to go back to the mid-90s and be out front on eCommerce.
For everyone outside the direct tech industry, you should be prepping your people and your systems for this new future. Mandate AI usage, so your company can start to understand and get familiar with this new paradigm of computing. Dedicate time to brainstorming use cases and experimenting with it. Designate someone in the organization responsible for reporting on the progress of AI adoption in the enterprise.
Where are the opportunities?
For companies that want to take advantage of being on the ground floor, there are a lot of unsolved problems that represent big opportunities:
-
- Security in this ecosystem is nascent to non-existent. This will be a huge problem.
- Payments and common infrastructure needs to be built. There is a trillion dollar market that wants to move money through this new platform. Building those rails is a massive prize.
- eCommerce in general. Every platform or surface in the world ends up becoming a portal to some kind of sales. This won’t be any different.
- Rethinking SaaS delivery. What aspects of your service today can work inside this new ecosystem? Where can you get rid of traditional UI and serve customers in this new OS?
- Social connection. Right now, the AI experience is very 1:1 – you and the machine. Like anything, our desire as a species to connect will mean we’ll also find ways to build human-to-human connection into this new space.
There are probably a million more of these, but you get the point: an entire new universe is being hatched, and it’s powered by human language and augmented by superintelligent machines. The amount of things that need to be retooled for this space are nearly infinite.
How do you prepare?
We’re early on, and it’s not clear yet how this will take shape, but there are some clues that help us think about where to invest time now prepping for this new future.
Data is the new oil
It seems clear that a new form factor is emerging, one that’s powered mostly by language, and some augmenting UI. The one thing that is clear is that the data layer is still the same. At the end of the day, you have to get information into and out of this operating system, and because this is based on internet protocols, having solid APIs and data portability is a solid bet to make now.
Agentic use will cannibalize traditional web use
Search volume is already dropping as a result of more people turning to ChatGPT and the others to answer their questions. As more and more people use installed agents on their devices, they’ll rely less and less on traditional web browsing. This means your data needs to be structured to work in these spaces, using Model Context Protocol to provide an adapter into those agents, so that as your consumers make this transition, you’re ready.
AI takes skill to learn, invest in training now
Using this new type of computing is like learning any new modality of computing: it takes time to get good at it. We know that this is here, and we know that it will only grow. Now is the time to invest in yourself and your team to ensure you’re getting the reps needed to get comfortable and proficient with this new technology.
When LLMs emerged into the public a few years ago, they looked like fun chatbot toys, something that could help you create a recipe, finish a song, or give you advice. But, rather quickly, these products are turning into an entirely new type of operating system, one that’s powered by human language and has superintelligent reasoning built into the core. This will have massive impacts across everything, and those that lean into this idea now will be ready to take advantage of being first movers in what will be an enormous new market.
