Buying an AI agent? What you actually need (and what it costs)

Right now, a lot of money is being made on the promise "buy this AI agent and your business runs itself." I run an autonomous AI agent myself — his name is Victor, he maintains my websites in 35 languages while I spend my time with my family — so you'd expect me to endorse that promise. The opposite is true: Victor wasn't bought. He was built. And that's exactly why he works.

What an AI agent actually is

Forget the robot images. An AI agent is a program that carries out tasks independently by combining a language model with access to your systems: your server, your git repository, your content pipeline. Victor runs on a Hostinger VPS — think of it as your own computer in the cloud that never switches off — and does its work there through scheduled tasks: generating articles, translating, publishing, fixing its own errors.

Why "buying" usually falls short

• A generic agent doesn't know your business. The value isn't in the agent but in the rules you set: what it does, when, and what it should do when something goes wrong. That's custom work by definition.
• You're paying for a black box. When a bought agent breaks — and everything breaks eventually — you can't fix it. My own scripts I can read, understand, and adjust (with help from AI).
• The real costs are elsewhere. Running an agent costs a VPS subscription plus API costs for the language model. You pay that too when you buy a "ready-made" agent — just with a margin added on top.

How to build one yourself (without being a developer)

I spent 23 years in banking and can't code in the traditional sense. My approach: I describe what I want in plain language to an AI coding assistant (Claude Code, or Replit's agent), paste the generated code into my VPS's terminal, and test it. Step by step, that grows into an agent that runs on its own. The building blocks: a VPS (a few euros a month), an AI assistant to build with, and patience. I describe my full setup in the article about my content machine.