How I Think We’ll Interface With the Digital World in the Future

A prediction I have is that interfacing with the digital world is going to be different in the future. The way I think it's going to happen is that companies like OpenAI and Anthropic have a lot of the compute, and they have the top AI models. Now everyone is using their tools to build things, and they're also training on a lot of the same data that pretty much the entire software world is training on. Eventually, the hardware world is going to train on that too. So they could essentially just build everything. Right now, we're at a stage where, if you want a website or a particular app built, AI can produce the slop to build it.

But it still has to be fine-tuned and made better. If you tailor your prompts better, you can get higher-quality apps. The better you tailor them, the better the apps get. And, of course, this general intelligence is going to get better. There's a lot of innovation going into taste for AI, aesthetics, and front-end work. It's still a big problem, but I don't see it as an unsolvable one. When it does get solved, we're going to be at the point where apps can just be launched and software can be created easily. Everyone can have their own versions, but it's going to get expensive in terms of cost and electricity.

But if the companies at the forefront of AI create it, or build a framework so people can easily tailor what they want on the spot, then a truly agentic OS could be possible. You could have a model small enough and efficient enough to run on an edge computing device. That's essentially going to be the future of software. Now, the main thing you have to be careful about is data privacy and security. Those are going to have to be built in, solidified, and extremely robust, especially when we get into breakthroughs in quantum computing.

So once we get to the security aspect, and I think that's where a lot of true intelligence labor should go, you could have conflicting superintelligences fighting against each other. I think with modern security in general, defense may be easier than offense because defense is a community-driven thing. Everyone is going to be working together on defense, versus offense usually being a small group of actors. Whatever one intelligence finds as a way to attack something, I think another intelligence will patch it up. So I think this is going to be a solved problem in the future as well. Unless we find out and conclude that nothing is actually secure.

In that case, we'll just have to find different methods for sensitive data. But that's something for the future. Because software can be created on the fly and run on the edge, everyone, as long as they have the hardware and storage, could configure their own cloud cluster and their own data security. They could also rent storage if owning a larger cluster ends up being too expensive. But I think it's going to be more of a hybrid in the future, where things can be more localized. And there is definitely going to be AI fatigue.

Because software is getting so easy and simple, I think the interface is likely going to change too, and we're going to get into more wearable devices. Speech-to-text has gotten so good that you really don't need to type anymore, or at least not nearly as much. It's so much faster to use speech-to-text. So wearable devices with microphones, such as smart glasses, smart bands, rings, and things like that, are going to become more popular. Edge computing is going to happen on those. If you had smart glasses with a camera and speakers, like the Meta Ray-Bans, and they were a lot better, you could essentially do everything.

Now, you don't have the screen, so how do you solve that problem?

The Apple Vision Pro was a great step because you could augment your current reality with screens. But mechanically and logistically, there were a lot of issues with the device that prevented it from becoming commonplace. Once those issues are sorted out, it will be almost a no-brainer to have that type of simple device instead of always having the latest phone that you have to charge more often. We've essentially reached the peak of that technology, I believe. When it comes to interfacing with software and other services, you will have an AI agent on each end.

So say, for example, there is a Vercel agent. A customer's agent could interface with it and transfer data if both are following the same data protocols and security exchange keys. That would be ideal. The Vercel agent could then spin up the website using information from that person's edge agent. The guardrails would be important to make sure no agent is acting up. There's going to have to be guardrail verification. And if any agents do not follow protocol, those agents will be reported. There might even need to be agent brokers out there where you can get an AI agent from.

I'm not sure about that idea. It's not something I've touched on as much. But somehow we're going to have this computing on the edge, this intelligence on the edge. And the big thing about this is the AI fatigue that's going to come. People are gonna go on social media, they're gonna go online, and there's going to be so much AI everywhere. No one knows what's real. No one knows what's fake. That's where the fatigue is gonna come in. Everyone is gonna start building slop to the point where no one really cares anymore about that exact slop. And in that case, I think authenticity is going to be a lot more important.

People are going to want more in-person interactions and to do more things in person, away from AI. People are already trying to become healthier and live healthier lifestyles. They're already feeling the drag and burden of all the dopamine craze from online stuff and the internet. They're leaving social media platforms. Dating apps are crazy. I think people are going to want more authenticity in their lives. They're going to try to get away from screens, software, and all of this stuff. But we still need it. We still need to interact with it. And that is where edge intelligence is going to come in.

And I think the world, or at least I hope more of the world, will start to head in the direction where people are more personable, they're outside more, and they're interacting more. The work or online interactions can be handled agentically. There just need to be secure protocols and order for that to happen. In terms of digital maintenance, verification, and monitoring, there can be AI agents, or general intelligence, set up to monitor it.

Different parties can have their own variations of these agents. They can have their own guardrails or their own things that they're looking for, so they know things are being checked the way they want them to be checked. It's going to be almost like an API. Instead of an application programming interface, it's going to be an agentic gateway interface. Or it may be just an agentic gateway, whichever sounds better. Maybe there will be another, better term. Different countries can also have their own agentic resources that communicate and interact with each other. It's almost like how, back in the day, we used to do things through a broker.

Like, for example, with the stock market or real estate, we went through a stock broker or a real estate agent. We went through a doctor, like a primary physician, to get tests and those types of things. But now we have agents to go through. You could go through the agents, and there could still be a human in the loop as well. People love the human in the loop, but that human might have a slightly different role.

And going back to the whole authenticity thing, with humans having their own AI agent or multiple agents, I really do like the idea of Worldcoin. Basically, you use your biometrics to create a unique identity while staying anonymous, because I think anonymity is going to be important in the future. But I do think that, with Worldcoin, you should be able to produce multiple identities from it. That would be ideal. You could have an anonymous identity if you want, and you could also have a public identity that is verified and validated through the biometrics.

But no one should really be able to relate those two identities to each other, for security purposes. Some countries might pass laws where you have to be able to identify all linked identities. I think in a country where privacy is important, like in the U.S., that's probably not going to be the case, but who knows. Something like Worldcoin would then prove that, okay, someone real is actually interacting. It's not just some bot army doing this. So I think a global organization that runs Worldcoin together, or some similar concept, will be the way to go. It could be almost like how the internet is a collaborative effort by the world.

Another direction we could go is that people get completely lost in the whole digital world, which is already happening right now. Some people might prefer that in the future, but I don't think it's gonna be as long-lasting. They might get caught up in the gaming space. Virtual reality and gaming are gonna become way more immersive and way more enticing in terms of the things that give people dopamine. Even things like pornography addiction might get worse when people can generate things based on their exact fantasies, giving them a higher and higher dopamine rush.

And especially when robotics comes along, some people who don't truly want to reproduce may feel like they don't need a partner. Instead of a human partner, they could get a robot that is able to completely outperform a human in the things they care about. Of course, that's definitely one potential direction. But I think, and I hope, more people go toward the opposite and try to become more real after getting fatigued by all of it. I still feel like there are gonna be things like video games, with more immersive versions coming out, and people are going to get more involved in those.

And that's fine, that's high-quality, good entertainment.

When it comes to interpersonal relationships, emotional intelligence develops and is expressed differently in different cultures. So that also gets a little tricky, and that's another reason why you might need different AIs trained in different knowledge scopes. For example, there could be obscure information in an Indian village or an Indonesian village that AI is not going to know because that information just isn't available.

So how do you go learn that information and bring it to AI?

So that's another job. Another job would be to explore, learn, and document. It would be about gathering the knowledge. The archaeologists, the knowledge people, whatever those roles end up being called. Those roles are going to become more important because there's so much knowledge out there that needs to be collected. It also needs to be truly understood, because that's not something AI and robotics can do entirely by themselves. They might be able to help out, sure, but there's going to have to be more to it.

And when it comes to interpersonal communication, there are facial expressions, observations, and things that still rely on our senses. AI or robots are not nearly going to be able to pick up all of that because they don't have the biological functions. We could say that, yeah, we could give them the biological functions. But a true human is probably going to react differently to that. And there's going to be a separation there.

People are already spending too much time on their devices. That's not as healthy, and there's a lot of fatigue from that too. Things like ordering specific items, checking email to see if there's anything interesting, filtering out spam, reading through work emails, and then writing messages can take so much time. For some people, it becomes fatiguing. And what we've seen with something like OpenAI is that a lot of people enjoy telling their AI to just go ahead and do these things. Once you do that, you realize that a lot of these tasks don't need a visual interface.

You can choose to look at it if you want to, but there's enough inference that the AI can often do it correctly. Then you're only shown the things you need to see. Some people might be hesitant to get into this type of tool because they feel like it's not good enough or it makes too many mistakes. There are issues with it right now. But I think in the future, once it gets better, that's the direction people are ultimately going to go in: toward this type of agentic interface, where they communicate with the agent to get things done.

And we're going to offload a lot of the tasks we currently do on our mobile devices. We don't need to personally do so many of these tasks. We're glued to our devices because now everything is an app. Everything is split into separate apps, and we're going to separate platforms to do separate things. But once it's all brought together into a single digital platform, we can just interact with an agent, and that agent can interact with the other apps. If the direction we go in isn't one universal platform, that's fine. We're still interacting with one thing, and that one thing is interacting with everything else.

And I think that's another good reason why we're going to interface with agents more instead of dealing with all of this stuff ourselves. It's another way to get away from the app fatigue and screen fatigue we have going on.

Everyone is now creating apps that bring all the other apps together. Like, there's an app that brings in your terminal, web browser, ChatGPT, plots, and code, everything together in one master app. A lot of people are also offloading to the cloud for compute so they can do things in parallel. And the operating system basically doesn't matter anymore because the interface is so customizable. So I think the differences in platforms and compatibility are slowly going to go away as more people execute things in containers. AI agents are just going to find the appropriate container where something can execute, be compatible, and work.

And then, if there needs to be backwards compatibility, they might develop an adapter on the fly. A bunch of things will just be put up and developed like that. So I think what is going to happen in the future is that we'll just have raw hardware optimized for compute. Optimizing the hardware is going to be really important. Then we just need really good software with a lot of pre-built components, or components that AI is already building and sharing online. There could also be a giant hub of software. GitHub is kind of like that right now, but there's still a difference and a learning curve because GitHub just has the raw code.

It doesn't have the raw software. There are software libraries out there, but you still have to take the code from GitHub, download it, install it, and use it. And maybe that's okay, because an AI agent can go on GitHub, find what it needs, install it in a container or small VM, virtualize it, do what it has to do, and then return the result. Or there could be containers that already have these things set up. The usage could be ephemeral. So, say there's a container that already has something like Photoshop on it. You clone it, use it, and then stop using it. Then another person can use that same Photoshop compute.

But there would be no memory of the previous person who used that Photoshop compute. It might be that type of shared computing style. That could become really important. Initially, I was thinking there might just be shared software, with a giant software library pool that people can pull from. But that kind of pool would only work if everything were highly vetted, scanned, and checked for security. There would need to be a bunch of AI agents doing that, along with people manually checking, validating, and verifying software before it gets approved. It could become a resource that agents go to for tools they are able to use and execute.

And it makes the most sense for these software repositories to be hosted by the people who own the agents, like OpenAI, Anthropic, or xAI. It might be a little trickier for the open-source ones, but they could use a community-vetted pool of tools. There could be a lot of bare-bones things in there. Then, from those, an easy customization or UI layer could be made. That could be one way this happens in the future.

Right now, Chinese LLMs like Kimi and GLM are top-tier and pretty close to the American frontier models, but they're way cheaper, and some of them are also open-weight. I think Kimi is open-weight, or maybe I'm thinking of Qwen. I would have to check the exact model. So basically, all that's gonna matter is: do you have the hardware, and can you pay for the electricity to run these AIs? Wherever electricity is cheaper, people might want to gravitate there, or that's where this is gonna be the most valuable. Because you can just get the hardware.

Right now, hardware is scarce. But once China comes out with really good, cheap hardware, or someone has a breakthrough in hardware, that hardware problem is gonna be completely solved. Then what are people gonna do? The value of Anthropic and OpenAI might go down, or it might not. But the whole point is that the whole SaaS system is gonna go down. Because right now, I'm seeing people point Fable or Kimi at a UI. There was an example with one particular application. I won't say what application it was, but basically it copied the app UI of Flighty. And it did it easily.

So we're getting to the point where we're able to copy a UI, or even copy an entire application, pretty easily. The value of just creating the software isn't there in the same way anymore. These developers are making tools and selling them to other developers. That doesn't make sense to me in a lot of cases, because those developers can just spin up their own tools. Or why not have the tool be open source? That would save token usage, so people don't have to go and create the same thing themselves.

Now, I know not everyone is gonna be a developer. But eventually these systems are gonna get so easy. Right now, you're able to go from speech to code. Eventually, you're gonna be able to go from speech to application. Once that is truly achieved, business owners and the people who actually want to use the software can create what they need. There are probably still going to be SaaS companies because of reputation and reliability. But a lot of companies are paying huge amounts for these tools. I know a company that was paying over a hundred thousand just for Slack. That's not gonna happen anymore. People aren't going to keep paying that much for Slack.

People are gonna have their own internal messaging systems, and they can configure their own security and protect themselves that way. They can create the software, then have another agent or a set of agents manage and host it. It's all going to exist in some kind of agentic software-hosting society or platform. And I'm curious whether there's gonna be a term for that. Maybe agents as a service. That could actually be a thing, although the acronym sounds horrible. It would be AAS. That's kind of funny, actually. It might be worth making fun of that in the article. But it could also just be a giant agentic playground. That might actually be better.

I'm not sure. Another point is that, if hardware becomes cheap, people can get their own hardware and start hosting their own AI. Then people can truly have a system where my AI talks to your AI, or my agent talks to your agent. That's essentially what the system is gonna be.