In December 2025, I closed my restaurant. Eight years. Michelin-recommended. Over a hundred people working across three locations at peak. I wasn't a chef — I was an operator. I ran the business.
The decision to close was mine, and it was right. But it left me with a question I'd been deferring for years: what do I actually build next?
The Call
I'd been watching AI tools the way operators watch food trends — skeptically. Every few months a new "AI-powered" product promised to transform restaurant operations. Most were chatbots with better interfaces, not fundamentally different capabilities.
But something shifted in late 2024 and into 2025. The products coming out weren't chatbots anymore. They were systems that could actually act — take reservations, route support tickets, monitor systems, manage communications. For the first time, I started paying serious attention.
A friend mentioned OpenClaw. "It's like having a personal assistant that runs on a computer you own," they said. "You can set it up to handle all your customer messages, your booking requests, your email — everything."
I started reading.
First Encounter With OpenClaw
The GitHub repository for OpenClaw had 270,000+ stars when I found it. That's an extraordinary number — it meant tens of thousands of technically sophisticated people found this valuable enough to endorse.
I read the documentation. The concept was exactly what I'd been looking for: a personal AI agent that runs on your own server, connects to your tools, and acts autonomously. Not a SaaS product where your data lives on someone else's servers. Your server. Your data. Your agent.
I signed up for Hetzner. I rented a server. I started following the installation guide.
Three days later, I had a running OpenClaw instance.
It also had no security configuration. I'd unknowingly left my agent publicly accessible to the internet. I found this out not because something bad happened — I found out because I stumbled onto a SecurityScorecard report about exposed AI agents and recognized the default configuration they described.
I fixed the security issues. That took another day.
Then I started trying to connect my tools.
The 70% Problem
Here's something that doesn't come through in the marketing for AI products: the demo is the 100% scenario. The product works perfectly in controlled conditions with a developer who knows exactly what they're doing.
Reality is the 70% scenario. You get most of the way there. You can see it working. The potential is obvious. And then you hit something that requires knowledge you don't have — SSH configuration, YAML syntax, Docker networking — and you're stuck.
I got to 70% with OpenClaw. The agent was running. It could receive messages. I had a couple of basic skills installed. But getting my email connected required API access configuration I couldn't get right. The Telegram integration kept dropping connection. One skill I installed broke two others.
I spent hours on Reddit and Discord. People were helpful. But the solutions assumed knowledge I didn't have. "Just modify the env file" — fine, which one, and what exactly do I change?
The documentation was written for developers. That's not a criticism — that's just what it was. OpenClaw was built by developers, documented by developers, for an audience that skewed heavily developer. I wasn't that audience.
What I Learned
The 70% problem isn't unique to OpenClaw. It's structural to how most technical products are built.
When a developer builds a product, they write documentation for themselves — for people who share their background, vocabulary, and instincts. The gap between "I know what SSH is" and "I've never opened a terminal" is enormous, and most technical products don't bridge it.
What I found after months in the AI agent space:
The technical complexity is real, but not permanent. The skills to run OpenClaw can be learned. The barrier is time and initial guidance, not fundamental difficulty.
Security is the biggest hidden cost. Most guides don't emphasize this enough. A misconfigured AI agent with access to your email and tools is a serious liability. Getting security right requires knowledge that isn't in the standard setup guide.
Maintenance is a second product. Installing OpenClaw is one project. Keeping it running — applying updates, monitoring for failures, managing skill compatibility — is a different ongoing project. Most users don't plan for this.
The power is real when it works. The agents I eventually got running changed how I worked. Email triage I'd spent 45 minutes on every morning took 5 minutes. Booking management became background noise instead of constant interruption. The technology delivers on its promise when the setup is done right.
Why Non-Technical Builders Have an Advantage
This sounds counterintuitive, but I believe it.
Non-technical founders see the use case clearly. We see the business problem — the 47 emails that need processing, the booking flow that loses customers at 3am, the monitoring that nobody does because it requires constant attention. We know exactly what a working agent should accomplish.
Technical founders often optimize for the elegant solution. Non-technical founders optimize for the outcome. These are different things, and the difference matters when you're building products.
I couldn't build OpenClaw. But I understood, more clearly than most developers, what the person who needs OpenClaw-style capability but can't set it up actually needs. Not better documentation. Not a friendlier CLI. A service that does the entire thing.
What Volos Is
Volos is the service I wanted when I was trying to get OpenClaw running.
You describe what you need — your tools, your workflows, the things that eat your time. Volos provisions a server, installs and configures your agent, hardens the security, connects your channels, and stays to maintain it. The agent is yours — on your dedicated Hetzner server, with your data — but you don't touch the terminal.
That's it. The entire value proposition is bridging the gap between the power of personal AI agents and the non-technical operator who needs them.
I built it because I was the customer. I knew what the customer needed because I'd experienced exactly what the current market didn't provide.
The restaurant taught me that operations — the unglamorous work of making things run reliably — is the hardest part of any business and the most undervalued. AI agents, done right, are an operations product. They handle the reliable, repeatable work so you can focus on the things that actually require you.
That's what Volos is for.
