Hey y’all — what does an AI-native internet actually look like?

In last week’s issue I mentioned how OpenClaw feels like a new level of agency for AI.

Now, an LLM can have its own hands and feet — it can actually take action on your behalf, based on simple instructions, and use judgement to execute those instructions well.

The interest is obvious. OpenClaw rapidly became one of the most starred projects on GitHub of all time.

But what does this rise imply for the future of how we interact with our devices, or the way the internet is structured?

This week I’m diving in to explore what that future will look like…

There are two big shifts we’ll explore here — one is the internet and how it’s structured, and the other is our devices and how we’ll interact with it.

They’ll happen in parallel, inching forward as the other does, rather than one being a dramatic change that pulls the other ahead. It stands to reason that the more obvious shift will be our devices though, so I’ll start there.

We all basically treat our computers and phones like workstations or personal productivity tools.

But agents change this — if agents can reliably take actions for us, we can do more by thinking about our devices not as workstations but rather as command centers.

Not as personal productivity tools, even, but rather as hubs to manage our swarm of agents.

Imagine a hierarchy of agents all on your phone, or laptop — some operate as administrators in a kind of OS layer for the agent swarm that allocates computer based on need and priority, manages interactions between agents, and enforces global rules or preferences you set.

Beneath that are highly specialized agents that each execute a singular purpose— interacting with specific APIs, browsing the internet, retrieving files, editing your photos, etc. The list goes on.

These administrators and executors will likely all be hosted and run locally on your devices. Cloud-based agents open your devices and workflows up to additional security risks, and they’re lightweight enough that having even a ton of them run locally won’t strain your device’s performance.

Even if you’re skeptical, imagine for a second that this actually works.

If these agents work in harmony and can produce good outcomes that would take you more time if done yourself. There will be many more of them than there are of you — and you can realistically only do one thing at a time.

The way you interact with your device changes. It’s not about surfing the internet or even using the apps you do today. It's about managing the agents. Making sure they’re performing as you want, and then consuming the information they surface for you (both at your request and proactively).

If they work, your laptop and phone become devices to orchestrate and deploy agents — command center, not workstation.

The need for a keyboard, standardized desktop UI, and mouse all fade away when you can communicate with your device by voice and it can take action for you (reliably).

It’s remarkably close to what OpenAI has discussed with their plans for a hardware device. For now, they plan to pair their (keyboardless, mouseless, UI-less) device with your phone, but it’s reasonable to expect that plans for a future version decouple the two.

Then the device captures your intent, the agents digest what you want, and then go find what you want from the internet. But what does that internet look like?

If agents are the arbiters of what info from the internet makes it to the humans on the other side, how will sites on the internet reorient themselves to ensure they’re making it through the AI filter?

And will people even use the internet directly at all?

My position is that they will, and the internet will have a split experience: basically an AI-optimized backend with a human-optimized frontend.

Agents prioritize understanding and efficiency. They want a robust robots.txt file and to be able to find the right information as quickly as possible. PS — my team and I can help optimize your site for GEO and agentic use. Book time here to chat.

Human users prioritize engagement and joy. They want experiential design and visual storytelling like this.

If agents are able to effectively crawl sites and find what their human orchestrators are looking for, the purpose for a human actually visiting a website themselves leans more firmly into the category of getting a “feel” for the brand’s identity, or either one-off or recurring experiences.

Think brand activations, or games.

The need for social networks likely doesn’t change in this scenario — if anything, it increases. Humans can interact with each other online more if their agents are doing the grunt work in the background.

My call to actions for y’all is to think about what you’d need to do to be ready for this future. It’s coming faster than we expect — OpenAI’s device is likely to launch this year.

And if you’re looking to start something new, there’s tons of ammo here aside from just an agent for some application-layer task. Build the administrators as they’ll be deeply embedded in workflows, retention will be higher if done right. Build agencies to help companies move into this world. Get ahead of it one way or the other.

A last note: one nice piece of this future is that accessibility of the internet will continue to rise. Humans who have impairments can rely on their agents to bring them the same information as quickly as anyone else is able to get it.

Advertise in my newsletter to get in front of 60,000+ founders.