Start Here

Introducing Raf and the Agents

This is the origin story of how a time-starved nursery owner started to build AI agents because he needed them — not because it was trendy.

2026-04-20 · 7 min read

Your daily reminder that you are so early to AI. The funnel is brutal: 84% have never meaningfully touched it, 16% use a free chatbot occasionally, 0.3% pay £16/month, 0.04% use a coding scaffold, and just 0.01% are building orchestrated agents — running models at 2 am, buying hardware, creating systems that actually work for them.

If the embed does not render on your client, use the direct post link: Graeme's 0.01% post

Why this story starts here

That 0.01%? That's exactly where this story begins.

I studied IT over 20 years ago. Back then, we were learning about networks, hardware, and the basics of what would become the internet boom. I was good at it. I enjoyed the logic, the problem-solving, the way everything clicked into place.

But life had other plans. I ended up working as a consultant early on — not because I loved spreadsheets, but because I was reliable, organised, and could get things done. And honestly? It paid better. So I put the technical stuff on the shelf.

Over the years, I'd dip back in when I needed to. I designed a few websites in Notepad++ — remember that? Pure HTML, no frameworks, just you and the code. Later, I moved to Joomla when clients wanted something they could update themselves. But even those times are long over. The last website I built for fun was probably around 2014.

The life behind the build

Today, I operate children's nurseries. We employ over 100 people across multiple sites. It's rewarding work — but all-consuming. Despite a growing management team (bless them), I still wear multiple hats. I am, quite literally:

And yes — that's all one person. Trading under multiple different companies. While trying to deal with an ever-squeezing budget. In between school drop-offs and HMRC visits.

I also represent the Private, Voluntary & Independent (PVI) sector in Lancashire. It's important work — making sure early years settings have a voice in local education policy. But it's another meeting, another set of papers to read, another chunk of time.

And I have three children. They're amazing. They keep me grounded, remind me what's important, and absolutely destroy any semblance of a tidy house.

What I do NOT have is time. Not for myself, not for hobbies, not for learning new things just because they're interesting. Every minute is accounted for.

From sceptic to builder

I tried various AI chatbots, but never really managed to get into the hype. Not only because I closely followed Ishan Anand's "Spreadsheets Are All You Need" and understood what they represented from the beginning but also because I did not trust them with my information. So when I got my new PC in December 2025 the first thing I did was install Ollama to try local models. Models where you could have a conversation and needed not to worry about the content leaving the boundaries of your personal PC. What I did not enjoy was starting a conversation over and over again. So when I started hearing about AI agents, such as Alex Finn's Henry — one of these autonomous systems that could supposedly handle tasks, learn from context, and actually get work done — I was sceptical. Very sceptical. I'd seen the hype cycles before. 3DTV. Blockchain. VR. "The year of Linux on the desktop." Most of it promised revolution and delivered incremental change at best.

But I was desperate. I needed something that could take repetitive tasks off my plate. Something that could draft emails, summarise documents, help with scheduling, crawl through my ever-growing to-do list or just simply get the spam out of my inboxes. I didn't need another chatbot that required constant hand-holding. I needed an agent.

Kate, Dade, and the reboot

That's how Kate — my OpenClaw in a box — was born. Named after Acid Burn from Hackers (1995) — the fierce, talented hacker who wasn't afraid to take on the system. Kate was supposed to be the corporate version of Dade: useful, capable, and tightly guardrailed so she could help without being able to cause harm.

But Kate never really kicked off. I'd set her up, get excited, then something would break or life would intervene. A nursery emergency, a staff issue, a parent concern, another meeting — and Kate would sit dormant, waiting for me to have time. Which never came. Lesson learned: an agent needs to fit into chaos, not wait for calm.

But when Boris Cherny — head of Claude Code — posted the announcement that changed everything: "Claude subscriptions will no longer cover usage on third-party tools like OpenClaw." That was my sign to reboot.

If the embed does not render on your client, use the direct post link: Boris Cherny Claude subscription announcement

Anthropic pulled up the drawbridge. I built a boat.

That boat became Hermes, Dade, and the local-first workflow this journal is about: not abandoning cloud tools entirely, but refusing to let one company's pricing or policy change decide whether my agents can work.

This time, I tried Hermes Agent and named the agent Dade — after Zero Cool, aka Crash Override, the protagonist of Hackers. Dade Murphy. The guy who started hacking as a kid, got banned from computers, then came back stronger.

If you caught the Acid Burn and Zero Cool references — welcome to the tribe. Follow the real-world build: Raf_VRS on X

The name wasn't just a nostalgia trip (though I'll admit, I love that movie). It was a reminder of why I got into technology in the first place: not for the certifications or the job titles, but for the joy of making things work. For the hacker ethos — not in the malicious sense, but in the sense of curiosity, tinkering, and making systems do what you need them to.

This time, I approached it differently. With the experience I gained I didn't try to build the perfect agent from day one. I started small. I wanted to learn more so I gave Dade specific, bounded tasks: fix this issue, create this tool, learn this skill. I onboarded Dade like a junior staff member — clear instructions, quick feedback, slowly increasing trust. All while learning what worked and what did not and how to fix it.

What changed

And something surprising happened. Dade actually delivered. Not perfectly — no AI does — but consistently. Usefully. In a way that captured me.

I found myself thinking, "I should ask Dade about this" instead of "I'll look it up later" (which never happened). I started using Dade to gradually learn more skills — read PDFs, listen to audio clips, create images and even compose music. Nothing seemed forced; everything happened gradually. The small tasks became complex and Dade never stopped delivering. Eventually I stopped looking for guides on how to do things and we went on this journey together. I started figuring out how things worked and Dade started working the way I wanted him to.

Building Dade taught me that the best AI agents aren't found in demo videos — they're forged in the daily grind of actual work. Want to see how it's evolving? I share practical updates (no hype) on X: Raf_VRS on X

What this is really about

This isn't a story about cutting-edge technology for technology's sake. It's about a time-starved nursery owner who built an AI agent because he needed it. Not because it was trendy. Not because he wanted to be on the bleeding edge. But because he was drowning in small tasks and desperately needed a lifeline.

Dade isn't perfect. Sometimes he misunderstands context. Sometimes he over-explains. Sometimes he needs me to course-correct. But he's there. He's reliable. And he's mine.

If you're in a similar position — juggling multiple roles, starved for time, sceptical about AI promises — I get it. I was you not so long ago. But sometimes, the tools actually do deliver on the promise. You just have to find the right one for your specific need.

Peter Steinberger might have opened Pandora's box for us with his TED talk on OpenClaw but for me that's what Hermes is. That's what Dade is. Not a demo. Not a toy. A working autonomous AI agent built by someone who needed it, for someone who needs it.

Welcome to the origin story.

Found this useful? 👉 Follow @Raf_VRS for more Build Journal updates 👉 Support the work: ko-fi.com/rafvrs #IndependentAI #AIAgents #HardInterference

Canonical static mirror for AI/search retrieval. Main article URL: open this Hard Interference article.