I Gave an AI Agent Its Own Computer and It Changed How I See the Future of SaaS

I Gave an AI Agent Its Own Computer and It Changed How I See the Future of SaaS

0:00

/293.067792

1×

A few weeks ago I came across a post on Reddit that explained how to set up a fully autonomous AI agent that simply links to your existing Claude subscription. No paid orchestrator, no new platform to learn. Just a Claude Code instance running 24/7 on its own virtual machine, remote-controlled from your phone.

I set it up the same evening in under 30 minutes and I have been refining it ever since. At our portfolio company Notehouse, it now runs regular BI check-ins, so I get a read on the numbers without opening our dashboards. It also researches new SEO opportunities and manages our outbound campaigns to report back to me, the kind of recurring work that is easy to glance over.

Here is how you can get this up and running in under 30 minutes and get a glimpse into what might be the future of SaaS.

The setup, step by step

The whole idea rests on one decision: give the AI agent its own machine with full access, then describe the outcome you want in as much detail as needed. Because it has full access to that machine, it does not wait for you to wire up an integration. If it needs a tool to finish a task, it will build the tool itself.

1. Spin up a small cloud server. I went with a €10 / mo Hetzner virtual machine, but any small VM works. Keep it separate from anything important, because you are about to give your agent full control of it.
2. Log in and install Claude Code. Now it's time to install Claude Code on your remote virtual machine and connect it with your existing subscription. Once done, you can then remote control it from your Claude desktop or mobile app.
3. Create a shared knowledge base. This is the important step for sharing knowledge with your AI agent. Markdown files are perfect for this: A shared Obsidian vault that syncs over Obsidian Git (free) or Obsidian Sync ($4 / mo) is a great fit. The vault should hold context about your business and how you like things done. The agent is instructed to regularly sync updates both ways.
4. Ready! Give it its first task to work on. Now that the setup is ready, I gave it access to the tools we are using at Waterglass to manage our business. Now that it has full context through the knowledge base and the right set of tools to work with, it's ready to work on its first task.

That is the whole system.

A small virtual server, your existing Claude Code subscription, a shared knowledge base and 30 minutes of your time. That is all you need to get started.

A glimpse of what comes after SaaS

There is a lot of writing right now about AI agents and what they will do to the broader SaaS market. Most of it is speculation, but one thing is absolutely certain by now: SaaS is changing and autonomous agents that can work independently on tasks for their users will shape the future of the software industry.

Only 30 minutes of setup showed me where the industry might be heading. As a founder and operator, this is what I would do next:

1. Build for headless. Agents will interact with your software business differently. They don't need a well-designed user interface, but rather a command-line tool or an MCP server to make interaction easier.
2. Your software business's moat is going to be data, compliance and deep integration. Easy-to-replicate software will become harder to defend against AI agents with clear instructions, a virtual machine with full access and lots of resources to build it from scratch.
3. Focus on distribution. When software becomes a commodity, your real lever is the distribution you build around it. This can take many forms: brand, content, SEO... you name it.

While the setup above still carries some complexity, I'm convinced that creating personalized AI agents to help us run our businesses will only get easier from here – and maybe change it for the better.

What recurring task in your business will you hand to your AI agent first?