Have you heard about AI agents yet?

No, they’re not just chatbots that know more. You could say they’re (chat)bots with a wider range of tools that can make their own decisions about how and what to do.

Like most people, I have a family and a regular job that takes up most of my time. Without AI agents, I wouldn’t be able to write this newsletter every week, post on LinkedIn, or build websites when I need them.

There simply aren’t enough hours in the day to do everything manually and still have time for things that matter to me.

It’s fascinating to watch how smaller companies that started using agents a while ago are now outpacing much larger competitors. They’re not working more hours; they’ve just stopped wasting time on things that AI can do for them.

What Makes an Agent Different

Since you’re reading this newsletter, I know you use AI here and there. 🙂 ChatGPT/Claude/Gemini… for research, help, writing… That’s great, but you’re still doing the work. You ask, it answers, you copy, you move on.

An agent is different.

You give it a goal. It figures out the steps. It uses your tools. It checks its work. It repeats until it’s done. You just review the result (or not even that).

You can imagine like this:

ChatGPT is a smart assistant you talk with. An agent is someone you delegate to.

The difference is small but enormous. Instead of “write me an email,” it’s “handle new leads from the website.” Instead of “summarize this document,” it’s “when someone is late with payment, send them a reminder and update my spreadsheet.”

You’re not giving instructions anymore. You’re just stating what you want to get in the end.

Use Cases

Sales & Leads

Lead Response Agent — A new lead (potential customer) fills out your form. The agent responds within minutes, personalizes the message, suggests meeting times, adds the contact to your CRM (customer relationship management system). All within five minutes of the lead arriving on your site.

Customer Support

Frontline Support — A customer asks a question. The agent checks your help documents, writes a response, sends it if confidence is high. If not, it directs it to you with context already attached.
Returns Helper — A customer wants a refund. The agent collects order details, checks your policy, writes a solution and puts it in queue for your approval.

Marketing & Content

Content Repurposing Agent — You publish a long post. The agent turns it into a blog draft and three LinkedIn posts. It puts them in a folder for you to review and customize.
Review Watcher — A new review appears on Google or social media. The agent summarizes the sentiment, writes a thank-you or apology and sends it to you for approval.

Operations & Administration

Email Triage — The agent reviews your email every morning, labels by urgency, writes responses to simple ones, and builds you a three-item action list for the day.
Collections Agent — An invoice is overdue. The agent writes a friendly reminder, attaches the invoice, sends it, and updates your tracking sheet when payment arrives.

Leadership & Reporting

Friday Snapshot — End of the week. The agent pulls stalled tasks, flags risks, grabs key metrics and sends you a summary that you can read in 60 seconds.

How Do We Build AI Agents?

There are quite a few tools. The most popular ones are tools where you drag and connect blocks. In each block, you define what happens at that point.

Here are the most popular platforms:

n8n Pros: You can self-host it, huge library of integrations, powerful for complex workflows, excellent for automation running in the background. Cons: Steeper learning curve.

Zapier Pros: Easiest to get started with, tons of pre-built connectors, no technical skills needed. Cons: Gets expensive quickly, limited AI capabilities, less flexible for complex logic.

Make.com Pros: Similar to the more popular Zapier, good balance of power and simplicity, affordable. Cons: Smaller community than Zapier, fewer templates.

OpenAI Agent Kit It was just “born” this week, when the owner - OpenAI (which owns ChatGPT) informed us about it. I quickly tested it and it still has some teething problems. It’s not reliable yet but worth keeping an eye on.

IMPORTANT

Agents are great until one sends the wrong email to 500 of your customers. 😬 Trust me, it happens.

But you can avoid such problems by following these guidelines:

Give it only what it really needs — Why would your email response agent need access to your entire accounting system? Give it only the tools it actually needs.

Check before it sends — For sensitive matters (refunds, mass emails, data exports), have the agent show you what it plans to do first. You just click “send” or “no”.

Log everything — When something goes wrong, you’ll want to know exactly what happened and when.

Have it ask rather than guess — If the agent isn’t 100% certain, it should ask you rather than do something stupid.

Start with one thing — First create one workflow that works. When that’s working, add the next one.

Here’s how I think about using agents myself:

I don’t create agents for everything. That would be a complete waste of time.

First I do it manually - I use ChatGPT or Claude to help me, but I still click and send things myself. This way I can see where the problems are.
If the same thing repeats several times a week - That’s when I start thinking about automation. I write down the steps, improve the prompts, test different approaches.
Then I improve the process - Whenever I see something that could be done better, I make a note of it. I gradually improve the process.
When it works without me - Only when I’m confident that the process works without my supervision do I turn it into a real agent that runs by itself.

That’s the essence.

Never automate something you haven’t done manually at least 5-10 times. Otherwise, you’ll just make mistakes faster.

In Conclusion

Agents won’t replace you. But they will ensure you don’t waste time on repetitive actions.

You’ll have more time for more important tasks. 😉

Talk to you soon, Primož

Try this: Think of one thing you do every week that annoys you. Take 30 minutes and see if there’s an AI tool that could do it for you instead. Don’t start setting anything up yet, just explore the possibilities.