Hi
I’m writing this from Paris on a Monday evening.
For the last four months, our team has been almost completely underground building Iris Labx. Long days, late nights, hundreds of product decisions, thousands of conversations with models, APIs, tools, and systems all circling one question:
What does software look like when the primary user is no longer a human?
We’re finally ready to let a small number of people into what we’ve been building.
Not a public launch. No press, no waitlist, no “look how perfect everything is” moment. This is a note to people who are early enough to see the shift before the rest of the market catches up and I’d rather you see it rough and real than polished and late.
Because I genuinely believe we’re entering one of the biggest platform transitions since the creation of the internet itself.
I was born in 1998. By the time I could read, the internet had already rewritten the world. It changed how an entire generation learned, worked, shopped, met people, started companies, and found anything at all. A kid born that year never knew a world where you couldn’t look something up, message anyone, or build a business from a bedroom. The web didn’t just add a tool it changed the default way humans reach the rest of the world.
I think the agent web does the same thing for the next generation. The kids being born this year will never know a world where you manually hunt for a service, compare ten tabs, fill out the same form for the fifth time, or sit on hold to book something. They’ll grow up expecting to state what they want and have it handled. How we discover services and tools, and how we interact with them, is about to change completely and it stays changed for decades. That’s the shift we’re building for.
For thirty years, software has been designed around human interaction. Buttons. Dashboards. Forms. Tabs. Search bars. Endless navigation.
Every interface assumes the same thing: a person is sitting in front of the screen. The entire stack frontend frameworks, SEO, UX, mobile apps, marketplaces, CRMs, SaaS was built on that one assumption.
It’s starting to break.
Millions of people already use ChatGPT, Claude, Gemini, Cursor, and Perplexity as their actual interface to the internet. And this is only the beginning. Very soon, most digital actions won’t start from a browser or an app. They’ll start from an agent:
“Find me the best flight.” “Book dinner for four near the office.” “Compare CRM vendors.” “Order groceries.” “Find an apartment.” “Run payroll.”Run my life. “Handle my accounting workflow.”
The interface is becoming conversational. The operator is becoming autonomous. And most companies are completely unprepared for it.
Every business today obsesses over being visible to humans, SEO, ads, social, App Store rankings, landing pages, funnels.
Almost nobody is thinking about discovery, visibility, and actionability for autonomous agents.
That’s the next battle. Because if an AI agent can’t discover your product, understand it, access it, transact with it, or operate it — then you quietly stop existing in the environments where decisions actually get made.
Not penalized. Invisible. The way a business with no website was invisible by 2005. The agents simply won’t see you, reach you, or transact with you — they’ll route around you to whoever they can.
Companies once had to become mobile-first, then cloud-first. The next transition is agent-first.
That infrastructure gap is what we became obsessed with. That obsession became Iris Labx.
We didn’t want to build “just another agent” We wanted to build the operational layer for the agent web four systems, one stack, each one making the next possible.
Think of Agentizer as the plumbing and wiring of the agent web the layer nobody sees but everything runs on.
Most software was never designed for agents. A huge share of the internet still demands manual navigation, brittle UI clicks, human login flows, hidden workflows. That’s friction an autonomous system can’t push through.
Agentizer turns existing software into something agents can actually call: APIs, MCP endpoints, CLIs, structured actions, agentfronts.
For services that aren’t agent-native yet no MCP, no CLI, nothing our agent falls back to computer use, driving the human interface the way a person would. That’s the bridge for today. But computer use isn’t the end state. The destination is a web where every product, service, and business is intentionally built for autonomous interaction from day one the same way companies once optimized for browsers and mobile.
Agentizer makes the web callable.
AgentNet is like Google except it doesn’t stop at finding.
Google indexes pages so humans can find them, then leaves you to click through and do the work yourself. AgentNet indexes tools, services, and businesses so agents can both find them and act on them. Every entry carries what an agent needs to actually operate it — not just a link to go look at. It’s a search engine where the results are doors, not directions.
Here’s the difference in practice. Take a restaurant in the Mission:
On Google, an agent finds a webpage hours, a phone number, maybe a photo. To book, it has to call, or squint through a reservation widget built for human fingers.
On AgentNet, the same restaurant shows up as something an agent can operate directly: it knows the place takes reservations, which times are open Friday night, that it seats parties up to six, and exactly how to hold the table. No phone call, no form. Find and book, in one motion.
Same shape everywhere:
a hotel → check real availability for your dates and reserve the room
a clinic → see open slots and schedule the appointment
a courier → get a live quote and dispatch the pickup
Today AgentNet indexes thousands of these APIs, SaaS platforms, businesses, commerce services, MCP servers, CLIs, workflows, operational endpoints. The index sits around 5,800 services right now and we’ll be 1 million at the end of this year. It should be hundreds of thousands, eventually millions.
The longer vision: agents will need their own equivalents of search engines, app stores, protocol registries, reputation systems, and transaction rails — built not for humans, but for autonomous systems.
If Agentizer makes the web callable, AgentNet makes it findable and usable.
This is the part I’m personally most excited about. LiM is the brain the translator between human and machine.
Most models today are optimized to generate language. But people don’t think in API calls. They think in intentions.
“Help me organize my trip.” “Find somewhere quiet.” “Get this done.”
The way a GPS turns “get me home” into turn-by-turn directions, LiM turns messy human intent — “book me a quiet Italian place in the Mission for Friday at 8, table for two” — into the right sequence of concrete actions across the right tools. It handles reasoning, planning, tool selection, workflow decomposition, memory, and execution.
The goal isn’t to generate text. The goal is to generate outcomes.
Iris is the best version of yourself but working for you you, tell it what you want, it figures out the rest. And it’s the part you can touch today.
Built on LiM, reaching into AgentNet, acting through Agentizer, Iris handles real tasks end to end: book travel, reserve restaurants, order services, coordinate multi-step workflows. Not a chatbot that hands you a list of links — an agent that actually does the thing.
The long-term vision is bigger than productivity. We think personal AI assistants become a new computing layer — an operational layer between humans and the internet itself. People stop using software and start managing intentions. Instead of opening twenty tabs and coordinating it all by hand, you delegate the outcome.
That single shift changes everything.
You tell Iris: “Book me a quiet Italian place in the Mission for Friday at 8, table for two.”
LiM reads the intent — Italian, Mission, Friday 8pm, party of two, quiet.
AgentNet finds the matches that are actually bookable and checks who has the table.
Iris holds the reservation and confirms — no app, no phone call.
If the restaurant is already agent-native, this runs straight through its endpoint. If it isn’t yet, Agentizer steps in and the booking happens through the normal site, on the restaurant’s behalf.
Four products, one motion.
We call the whole shift the move from the World Wide Web to the World Wide Agents WWA. The human web needed its foundational infrastructure built around the time I was born. The agent web needs its now and that’s the decade we want to spend building it.
The honest truth is that this is still very early.
Things break. Latency is rough. Tool execution sometimes fails. Some workflows feel magical and others are frustrating. The whole ecosystem around us is still immature.
We shipped anyway, on purpose because infrastructure products can’t be perfected in isolation. You only learn from real-world behavior: what people try to delegate, where agents fail, where systems collapse, where trust breaks, where orchestration gets hard.
The early web was messy, fragmented, and unreliable too. But underneath the chaos was a fundamental shift in how the world would operate. That’s what this feels like right now.
So please — go break it.
The deeper belief behind Iris Labx is that agents are becoming first-class economic actors on the internet. They’ll search, negotiate, buy, sell, coordinate, schedule, optimize, and transact on behalf of humans and companies.
Once that happens, the infrastructure layer becomes everything: discovery, trust, execution, identity, coordination, interoperability.
Someone has to build the operating layer between autonomous systems and the real world. That’s the problem we want to spend the next decade on.
1. Try Iris. Give it something real not a benchmark prompt, not a toy request. An actual task you’d genuinely want delegated. Then tell us what worked, what failed, what felt magical, and what completely broke. The ugly feedback is the useful kind.
2. Forward this to one person. Not a crowd — one person. One founder, one operator, one engineer, one product mind who should already be thinking about what the internet looks like when agents become the primary interface.
3. Open a door. If your company is exploring agent infrastructure, commerce, payments, travel, productivity, marketplaces, operations, or autonomous workflows — I’d love to talk. → joel@irislabx.com
We’re still early. But some shifts only look obvious in hindsight.
I think this is one of them.
Joël, Cofounder, Iris Labx
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.