Automation and agents

This is where we go from talking to AI to letting it do things for us. An agent is an AI that doesn't just answer, it acts: it reads your email, moves files, launches tasks or looks after a customer. It's the most promising part of all this, and also the one that gives you the biggest scares. In this section I share what has happened to me working with agents and harnesses like Claude Code, DeepSeek Harness or OpenClaw: the good parts and what broke.

The test for a case to belong here is that the AI connects to other programs or works inside a workflow that keeps running without me sitting in front of it. If I'm just speeding up one of my own tasks with an assistant, it goes in Productivity; if what I'm building is a website or a utility, in Build your own tools.

I also talk about what you no longer need: a lot of things that used to mean building workflows with n8n or Make can now be asked of the AI in a single sentence. Not always, and I'll tell you when it still makes sense to build a workflow. And how to connect AI to your tools (email, files, calendar…) through MCP, without handing it access to everything without thinking.

And the part a business cares about most: customer service with chatbots, on your website or on WhatsApp. How to set one up, what it costs per month and, above all, what it does when it doesn't know the answer, because a chatbot that makes things up in front of a customer is worse than having none at all. I'm not selling total automation here: I tell you where it genuinely saves work and where you still need to keep an eye on it.