---
title: "How to learn to work with AI agents: start with a problem"
description: "Why I teach people to work with AI agents through one real problem of their own, not through prompts and tools, and what they built on their own afterwards."
url: https://promptityourself.com/topics/learning-starts-with-a-real-problem/
---

# Learning to work with AI agents starts with a real problem

Why I teach people to work with AI agents through one real problem of their own, not through prompts and tools, and what they built on their own afterwards.

 Mate Visky · 29 October 2026 · 6 min read

When I try to judge whether someone has really learned to work with an AI agent, I no longer look at how much we managed to build together during a session; I look at what they are able to do on their own in the weeks that follow. That shift in emphasis underpins Prompt It Yourself and explains why I teach the way I do.

Most AI courses begin with the technology itself: a list of prompts to memorise, a tour of features and a handful of tools the learner is told to master. I think this sequence is back to front. People learn faster, and retain more of what they learn, when they begin with a real problem from their own work or everyday life, because the problem gives every new technique a purpose and provides a way to test whether it has actually helped.

## Domain knowledge is the scarce resource

I have spent twenty-five years in IT, and the most consistent lesson I have drawn from that time is that people understand their own businesses far better than any outside expert ever will. They know their customers, their processes and the points at which things quietly go wrong. For much of my career, my job was to translate those problems into specifications that technical teams could use to develop solutions.

AI agents change that arrangement. Instead of translating someone's problem for a development team, I can now teach them to communicate with an agent directly: to explain the goal clearly, to ask questions that help define the problem more precisely, to challenge the answers they receive and to check that the result really works. In my view, this ability to communicate and collaborate with an agent matters far more than memorising prompts or keeping up with every new tool, and it is one of the most useful skills anyone can develop today.

![The new skill is not memorising prompts. It is explaining a real problem clearly to an AI agent, then checking what it does.](https://promptityourself.com/media/articles/new-way/01-skill.jpg)

The new skill is not memorising prompts. It is explaining a real problem clearly to an AI agent, then checking what it does.

## What happens when people start with their own problem

On my show, people bring a problem they want to solve; we install an AI agent on their own computer and work together on the problem. They type and make the decisions, while I guide them, suggest what might be possible and help them check the results. Our attempts do not always succeed at first, and those failures are part of the learning rather than a distraction from it.

My sister Bernadett, a psychologist in Budapest, had always needed someone else's help to make changes to her website and had never opened a terminal. In our first session, she got her own agent running, and two days later her bilingual website was hosted in her own account, where she could edit it herself. Adrienn, an accountant in Sydney, wanted to stop sorting signed tax returns into client folders by hand, so we connected her agent to the tools she already used and began automating the process. Istvan, a projection mapping artist, needed to find customers, and after four hours he had around 3,000 potential leads, organised according to who might need his work and when.

These are different people with different problems, but the same principle applies to all of them: begin with what you want to achieve, not with what the technology can do.

![Begin with one problem you understand, not with the list of tools that might solve it.](https://promptityourself.com/media/articles/new-way/02-problem-first.jpg)

Begin with one problem you understand, not with the list of tools that might solve it.

## You don't know what to ask yet

When you are new to something, the hardest part is often not finding the answer but knowing which question to ask, because you do not yet know what is possible. That is why I think out loud during a session: I explain why I ask a particular question, what I am trying, and what I do when the result is poor.

Halfway through Istvan's session, his agent delivered a piece of research on potential customers, and I dismissed it at once as poor work. Istvan laughed: I knew the capabilities, he said, so I judged it immediately, while he was "still at: wow, I got a research I could never do on my own." We were looking at the same result, but I knew what the agent could do, and he did not yet. So I said out loud what I wanted instead: the big events around the world for the next two years, from openings and concerts to new buildings, collected into a library of leads.

That habit is what I want people to take away: not my prompt, but the expectation that the agent can do better, and the confidence to say so. I don't expect anyone to learn in a few sessions what took me twenty-five years. I want to show enough of how I think that people become confident enough to explore without me.

## Building what you need instead of subscribing to more

I also question the assumption that every new problem calls for another software subscription. It is usually wiser to start with what you already have. Sometimes an existing service is the right answer, but sometimes your agent can develop a small solution that fits your needs more closely: an automation, a script, a website or a tool designed around your own workflow. Security, maintenance and cost still matter, and nobody should try to build everything themselves; the point is to understand the options before buying another product. The agent can help you build, but the decision about what is worth building remains yours.

## The most important results come afterwards

The clearest difference becomes apparent after the sessions end. A month later, Adrienn came back with a reconciliation tool she had built herself, which reduced a three-hour task to about twenty minutes. Bernadett returned with a client questionnaire she had completed on her own and later built her own CRM. Ildiko, in Sydney, had been working on a voice agent that checks in on elderly residents; a month after our session, she came back with the finished agent, complete with memory and automated reporting, and she had also set up a separate online shop.

None of them was waiting for me to solve their next problem. They were solving it themselves, and I consider that the real result of the work.

![The real result of a session is what people build on their own a month later.](https://promptityourself.com/media/articles/new-way/03-afterward.jpg)

The real result of a session is what people build on their own a month later.

## Knowing when to stop is part of the skill

Working well with AI agents also means not trusting them blindly. Denes had spent three months building a sustainability reporting application when we asked four different agents to challenge the business idea; the estimated cost of acquiring even one customer led him to reconsider, and he changed his plan. Peter, a former professional poker player, wanted to collect data from a betting website, but after his agent checked the site's terms and Peter read them himself, he decided not to build the bot.

Nothing was built in either case, yet something important happened: both of them made an informed decision. An agent can research, make suggestions, write code and make mistakes, which is precisely why the person using it has to retain responsibility for the decisions.

> The best outcome is not a person who comes back for more help; it is a person who no longer needs it.

## Why the best outcome is not needing me

I am aware that this is an unusual business model, but after many sessions I am convinced that the best outcome is not a person who comes back for more help; it is a person who no longer needs it. What I enjoy most is seeing what people create afterwards, especially the things we never discussed, because this shows that they have learned to turn to their own agent first when the next problem arises.

That, I believe, is what AI education should aim for. It should neither try to turn everyone into a developer nor simply provide hundreds of prompts to copy. It should teach people to communicate, think, question and work with AI agents using the knowledge they already have: to start with one problem they understand, explain it to their agent, explore the options, challenge the answers, build something small and continue from there. Most people are surprised by how far this process takes them.

The stories and full sessions of the people mentioned here are available at [promptityourself.com/stories/](https://promptityourself.com/stories/).

## Find more like this

### What it is

-   [Articles](https://promptityourself.com/tags/articles/)

## More guides

[All 10 guides](https://promptityourself.com/topics/)

[How do I make my own website without paying every month? Build a website on your own domain with an AI agent, and change it yourself whenever you want. 9 sessions](https://promptityourself.com/topics/website-building/)

[Connect Xero to the files I already have? Connect Xero to the tools you already pay for with an AI agent, and stop typing the same figures twice. 2 sessions](https://promptityourself.com/topics/accounting-integration/)

[File signed documents into the right client folder automatically? Move files between Dropbox and OneDrive with an AI agent. Signed documents file themselves, with no subscription. 2 sessions](https://promptityourself.com/topics/file-movement-integration/)

[How do I find companies that would buy from me? Find the companies that would buy from you, and build the lead list yourself with an AI agent instead of renting it. 2 sessions](https://promptityourself.com/topics/lead-generation/)

[Put a questionnaire on my website for free? Put your own client questionnaire on your website with an AI agent, so you know who is asking before you speak to them. 3 sessions](https://promptityourself.com/topics/survey-creation/)

[Make pictures for my business without paying for stock? Make pictures for your business with AI, on your own computer, in as many versions as you want. 5 sessions](https://promptityourself.com/topics/image-generation/)

[Cheap CRM for a one person business? Build your own simple CRM with an AI agent: one place for every client, lead and session, not four spreadsheets. 1 session](https://promptityourself.com/topics/client-records/)

[Get AI to review my own work and tell me what is wrong? Ask an AI agent to review your own work. It reads all of it and tells you what is wrong, at any hour. 1 session](https://promptityourself.com/topics/ask-it-what-is-wrong/)

[Run my AI agent on a machine at home? Move your AI agents to a home server, so they keep working when you shut the laptop.  1 session](https://promptityourself.com/topics/home-server-migration/)
