No one is AI-native

I’ve heard a lot of people use the term “AI-native” recently, usually to describe young people who use AI a lot, or companies that are built around AI, or developers who use Cursor and Claude Code and loops or whatever else for basically everything.

But I don’t think any of these people are AI-native.

In fact, I don’t think anyone is AI-native yet.

AI-first is an intention. AI-native is an inevitability.

“AI-first” means that when you have a problem, you think about solving it with AI first. You have to make that decision to use it. And then you have to know which tool to use, and to get something useful out of it.

This is actually a large part of my job. I work in AI Operations, and a lot of what I do is essentially helping people become more AI-first: showing them what AI can do, showing them where it can save them time, pushing them to use it for more ambitious things, getting them to default to AI first.

But the fact that this job exists is probably a pretty good sign that we are not AI-native.

Let’s compare AI to the internet.

There is no internet adoption team in my company. No one needs to convince people to use the internet more. No one is tracking whether employees used enough internet this quarter.

There are no presentations about becoming “internet-first.”

The internet is just there. And more importantly, all the systems around us assume that it is there and we’ve built up infrastructure to make it easier to use to get things done faster or better.

My bank assumes I use the internet. My work assumes I use the internet. My kids’ school assumes I use the internet. Government services increasingly assume I use the internet. When I travel, or pay for something, or I need directions, when I book a hotel, when I need to contact someone, the easiest way to do all of these things is online.

The strange thing would be to do them offline.

And that’s the big difference. AI is not there yet. We still have to push ourselves to use it, rather than to push ourselves to not use it.

We’re still “going” to AI

When I was younger, people used to say that they were “going online.”

That phrase sounds strange now because there is no real distinction between being online and not being online anymore. Your phone is online. Your TV is online. Your car is online. Your watch is online. Your doorbell might be online too.

You don’t go online anymore. You just are. We’re internet-native because being online is easier than being offline; every institution that we touch assumes we have access to it and is optimized for that; opting out of the internet causes a lot more headaches than it fixes.

But AI is still something that you go to. You open up your ChatGPT, or your Claude or Cursor. You deliberately invoke the AI. And if you want anything more sophisticated than that, you quickly discover that AI still requires an enormous amount of scaffolding around it.

I’ve been experimenting with this myself by building a family assistant.

The idea is extremely simple: I want an AI that knows our family calendar, knows what my daughters are doing during the week, understands appointments and pickups and scheduling conflicts, and that my wife and I can message normally.

This is exactly the kind of thing people imagine when they talk about the future of AI. And to be fair, the AI itself is not really the difficult part.

I needed a server and API keys. I needed to connect calendars. I needed a separate WhatsApp number. I needed permissions and authentication and a way to make sure the thing doesn’t accidentally reveal information to the wrong person. I have to think about maintenance, security, privacy, failures and what happens when some integration randomly stops working.

I enjoy doing this stuff, so I’m willing to spend hours setting it all up and testing it over and over again.

But this is obviously not native.

If society were actually AI-native, getting a family assistant should be boring. It should be something I can just switch on or opt into. The models are already smart enough that if we had the infrastructure to exploit them to their capabilities – today’s capabilities – we’d move faster towards being AI-native.

AI-native governments might subsidize one for elderly people, or give every child an AI tutor. Every employee gets some kind of persistent assistant the day they start a new job, the same way they currently get an email address and access to Slack.

Generic “AI” disappears

At that point, nobody would call any of this “AI.”

This is perhaps the part that interests me most: Technologies seem to become native when we stop noticing the technology itself. There was probably some period when electric lighting felt like technology. Now a light is just a light. No one says that their office is electricity-native.

The technology has disappeared into normal life, and I suspect AI will do the same thing.

You won’t ask someone in a job interview whether they “know how to use AI,” in the same way you don’t ask someone whether they know how to use the internet. You’ll ask whether they know how to set up their assistants for different environments, they’ll ask what the company’s knowledge system is, how it’s being exploited. AI itself will be assumed.

The same thing will probably happen to AI budgets.

Right now, companies still talk about how much money they should spend on AI. But internet access is not treated that way. No one looks at the internet bill and asks whether the marketing team really needs access this quarter.

It’s infrastructure. And that is ultimately the threshold. AI is going to become native when using it is easier than not using it, and when the institutions around us are built with the assumption that it exists and that it is valuable.

When a developer uses Claude Code for twelve hours a day, or startups claim to be “AI-native” on their homepage, or when a company reaches an 80% AI adoption target – those are signs that we’re still in the transition.

The amount of effort we are currently putting into becoming AI-native is, in a strange way, evidence that we aren’t.