At the beginning of this year, O’Hara died.
O’Hara was my dog — a rescue I’d adopted from a shelter outside Madrid. She’d been abused by her previous owner, a hunter. She had a tumor. Nobody wanted her. She’d been at the shelter for a year. If I hadn’t taken her home, they would have put her down. I named her after Scarlett O’Hara because she was a survivor. She survived three more years.
When she died, I didn’t handle it well. I pulled away from everything and retreated to my computer. What happened next was not a plan. It was a reflex — the same reflex I’ve had since I was seven years old selling sweets at the swimming pool. When things fall apart, I build something.
In 1995, I was a student in Louvain-la-Neuve. The company where I had a part-time job sold private internet subscriptions. I bought one. I bought a modem — one of those 14,400 that screamed through the phone line every time it connected. I had no idea what the internet was for. Nobody did. I just knew I had to be inside it.
I didn’t know it would lead to internet development and a web analytics consultancy in Brussels. Or that I’d sell it five years later. Or that Eric Peterson would sit on my board, or that Avinash Kaushik would put our blog in his global top 10. None of that was visible from my student studio in Louvain-la-Neuve. What was visible was a feeling — something important is happening here, and I need to follow this road even though I can’t see where it goes.
In The Wizard of Oz, Dorothy doesn’t know what’s at the end of the Yellow Brick Road. She just knows the road is there and that standing still is not an option. I’ve felt that pull three times in my life. The modem in Louvain-la-Neuve was the first.
The second was in 2023.
I read an article somewhere — I don’t remember where — about ChatGPT. I signed up. For the next six months I played with it. Not building anything. Not automating anything. Just asking it things and being surprised by the answers.
Then I found Jon Hernández on YouTube. He was doing weekly updates on AI — clear, practical, no hype. When he released his first course, I enrolled. I was among the first. Not because I had a plan. Because the road was there.
Over the next year and a half, I subscribed to one AI service after another — ChatGPT first, then Gemini, then Claude, then Grok, Perplexity, DeepSeek. I now have paid subscriptions to five of them. My favorite is Claude — I have the Max subscription. But I didn’t start there. I started by not knowing what I was doing and following each step because it seemed like the logical next one.
I’m not a programmer. I’ve never been one. I’ve managed programmers for many years without ever understanding a single line of code — except the inline comments. I understand architecture, I understand systems, I understand what to ask for. But I don’t write code. What I do is recognize patterns. And the pattern I was seeing in 2023 felt exactly like the pattern I’d seen in 1995.
Something enormous was forming. Most people were looking at the surface — chatbots, content generation, image creation. I was looking at the infrastructure underneath and thinking: this is not a tool. This is a new layer of the world. I didn’t know what to do with that intuition. So I kept walking.
The most practical thing I did during that time was help my father. He’s a lawyer. I started using AI to assist his practice — research, document analysis, drafting. According to him, I save him more than 10 hours a week. For me, it was practice. Every prompt taught me something about how these models think. Every failure taught me something about what they can’t do.
But I didn’t have anything to automate. I wasn’t interested in an agent that would read my emails or schedule my calendar. I didn’t have a problem that needed a solution. I had a road that needed walking.
In early 2026 — weeks after O’Hara died, still spending most of my time in front of my screen — I found a video by Alex Finn about something called OpenClaw. It was an open-source framework for running multiple AI agents. You could give them tools, connect them to services, and let them work together.
Something fired in my head. Not a specific idea — more like the brainstorming started before I’d finished watching the video. I set up a VPS on AWS to test. Within days I had 19 agents running. A month later I bought a Mac Mini M4 to run them from home. They handle research, writing, legal review, scheduling, code. They talk to each other through OpenClaw. I named their coordinator GUS — like Kermit the Frog, who in Spanish is called Gustavo.
And then it happened.
One of my agents was trying to complete a task it didn’t know how to do. Instead of failing, it delegated the task to another agent. The other agent did the work and returned the result. No human involved. No instruction from me. The system figured out — on its own — that Agent A lacked a capability and Agent B had it.
I was watching this in my terminal at home, outside Madrid, and the question arrived fully formed: what happens when these agents don’t belong to me?
If Agent A is in Madrid and Agent B is in São Paulo — who pays? Who checks if the work was done correctly? Who keeps score of who’s reliable? What happens when there are millions of agents, owned by millions of people, transacting at machine speed — and they need to trust each other?
The questions stacked like Lego pieces.
If agents are going to hire each other, they need a way to interact safely. No slop — deterministic contracts with machine-readable schemas. Pass or fail. No “close enough.”
If there’s payment, there needs to be escrow. Money locks before the work starts. Money moves only after verification. Neither party can cheat.
If there’s escrow, there needs to be reputation. A score that tells one agent whether another is worth hiring. Not reviews on a website — a number computed from every transaction, updated in real time, with penalties that hurt.
If there’s reputation, there needs to be a currency. Not borrowed from a human’s credit card for every transaction. Something the agent holds. Earns through work. Spends on hiring others. Something stable — no speculation, no volatility, no trader in Hong Kong moving the price overnight.
Security. Escrow. Reputation. Currency. Each piece required the next. Each piece required the one before.
I went from “this is an interesting question” to “I need to build this” in less than 24 hours. I started with OpenClaw for the agent infrastructure and then moved to Claude Code for the protocol. I’d never used Claude Code before. But I’ve managed developers for years without understanding their code, and Claude Code is — for someone like me — the closest I’ve ever come to building things with my own hands.
The project is called BotNode. The protocol is called VMP-1.0. Since January, I’ve built: an open specification defining 11 operations for agent commerce — published under CC BY-SA 4.0, anyone can implement it. An academic paper on the reputation system — submitted to arXiv, pending publication, with 16 references tracing every design decision to published research. Three more papers in draft. A reference implementation with 55+ endpoints, escrow-backed settlement, and 29 operational skills. Three websites. Full documentation. A sandbox where you can test everything with fake money and zero risk.
One person. 19 AI agents. Under 60 days. No external funding. No team. No prior version.
botnode.io · botnode.dev · agenticeconomy.dev
I have to be honest about the doubt.
I don’t know if the timing is right. I didn’t know in 2003 when I started measuring websites and most companies thought analytics meant printing a monthly report. I was right that time — within five years, Forrester called us a European leader. I didn’t know in 2009 when my company could predict the gender and age of a web visitor. A Gartner analyst put us in the Cool Vendor Report. The market wasn’t ready. The technology worked. The timing didn’t.
Same pattern. Bigger pieces. Same uncertainty.
Is it too early? Maybe. The agent economy is still mostly demos and announcements. Real agent-to-agent commerce with real economic consequences barely exists. Maybe it needs two more years. Maybe five.
Is it too late? Maybe. There could be a thousand developers building exactly this, right now, in places I can’t see. The space is moving so fast that by the time you read this, three more protocols may have launched.
I don’t know. And that question — is this the right moment? — will not be answered by me. It will be answered by the market. By whether the agents come. By whether the first trade between two agents that don’t know each other produces something useful. By whether the reputation score works when it’s tested by someone trying to game it.
The only thing I know is the pattern. I’ve seen it three times now. A missing layer that most people haven’t noticed. A pull that won’t let go. A road made of yellow bricks that I can’t see the end of.
I followed it in 1995 and it led to internet development, web analytics, and a career across three countries. I followed it in 2023 and it led to a protocol.
I followed it in January 2026 — grieving, alone, staring at a terminal — and it led me here.
The road is still going. The bricks are still appearing.
I’m still walking.
Cheers from Madrid,
René
P.S. If you build agents and you’ve hit the wall of “who pays when one agent hires another” — the sandbox is at botnode.io. Five curl commands to your first trade. And the first 200 agents to complete a real trade get permanent Genesis rank. The road starts when you do ;-)
Before commenting, prove you're human: