The internet's next users won't be human

Iris Labx — Founder letter, June 2026

Hi,

It's thursday and i'm writing this from my room.

For thirty years, every piece of software on the internet was built on one assumption: a human sits in front of a screen and operates it.

Search. Browse. Compare. Fill forms. Switch tabs. Copy-paste between systems.

That assumption is breaking fast.

Millions of people already use AI as their primary interface to information. They don't open websites; they describe what they want. Today, AI answers questions. Tomorrow, it completes tasks. And when the primary user of software stops being a human and becomes an agent acting on a human's behalf, the internet itself has to be rebuilt.

Companies once had to become digital first, then mobile-first and cloud-first. Today and next transition is agent-first and almost none of the infrastructure for it exists yet.

That's the company we're building. After four months heads-down, Iris is now open to a small group of invited users.

The shift: from access to information to access to action

The internet solved access to information. The agentnet solves access to action.

Instead of searching for a restaurant, an agent books it. Instead of comparing flights, an agent purchases the best option. Instead of navigating software manually, you describe an outcome:

“Book a quiet Italian restaurant near my hotel tomorrow at 8pm.” “Find an apartment in San Francisco under my budget.” “Run payroll.”

The future interface isn't a website. It's an intention. And the future user isn't always a human it's increasingly an agent acting for one.

The visibility crisis nobody is pricing in

Every company today optimizes for human discovery: SEO, ads, landing pages, app store rankings.

But when the buyer is an agent, none of that matters. If an AI system cannot discover your product, understand it, and execute actions through it, your company becomes invisible inside the environments where decisions are actually made not because your product is bad, but because agents literally cannot operate it.

Whoever builds the discovery and execution layer for agents sits at the toll booth of the next internet. That's the position we're going after.

What we've built

Four systems, one stack:

1. Agentizer — make the web callable. Most software was never designed for agents. Agentizer converts existing software APIs, MCP servers, CLIs, structured actions, agentfronts into something agents can operate. When a service isn't agent-native, it falls back to computer use and operates the interface the way a human would.

2. AgentNet — make the web discoverable. Google indexes webpages. AgentNet indexes actions: things an agent can actually do — book restaurants, reserve hotels, run workflows, purchase services, access software.

Today AgentNet indexes thousands of APIs, SaaS platforms, businesses, commerce services, MCP servers, CLIs, workflows, and operational endpoints. The index sits at roughly 5,800 services today and is growing rapidly. Over time, we expect it to scale to hundreds of thousands and eventually millions of callable services across the internet.

3. LiM — translate intention into outcome. Most models generate language. Humans think in intentions. LiM, our Large Intention Model translates human intent into executable plans: reasoning, tool selection, workflow decomposition, execution, memory, coordination. Its goal isn't generating text. It's generating outcomes.

4. Iris — the product you can use today. A personal AI agent built on LiM, AgentNet, and Agentizer. You tell Iris what you want; it figures out how to make it happen. The goal isn't to give you links. The goal is to finish the job.

What it looks like

Tell Iris: “Book me a quiet Italian restaurant in the Mission on Friday at 8pm for two.”

LiM understands the request. AgentNet finds restaurants that are actually bookable. Agentizer connects to the reservation system. Iris confirms the booking.

No forms. No phone calls. No tab-switching. One intention, one outcome.

Why now and why launch early

We're early, and we're honest about it. Some workflows are magical. Some break. Latency isn't where we want it.

We're shipping anyway, because infrastructure cannot be perfected in isolation. The only way to build the agent web is to put it in real hands and learn from reality. The early internet was messy too underneath the chaos was a platform shift. This feels the same.

The bet

Agents are becoming first-class economic actors on the internet. They will search, negotiate, purchase, coordinate, schedule, and transact on behalf of people and companies.

When that happens, an entirely new layer becomes necessary: discovery, execution, identity, trust, coordination, interoperability. Someone has to build it. That's the problem we want to spend the next decade solving and we believe the company that owns this layer will be one of the defining companies of the agent era.

Forward this to one founder, or operator who should be thinking about what the internet looks like when agents become the primary interface.

We're still early. But some platform shifts only look obvious in hindsight.

I think this is one of them.

Joël EtseCo-founder, Iris Labx