Can I build a business that pays me a salary?

The internet is full of people making thousands a month with artificial intelligence without breaking a sweat. I break a sweat choosing a phone plan, so I decided to test it on myself. This is the first part of a challenge: pick a business, actually set it up and share the numbers, however they turn out. The challenge is already over: I ran it from 1 June to 9 August 2026 and now I'm telling it in parts.

The question is simple: can I build a business that pays me a salary? Not in theory or in an optimistic spreadsheet, but for real, with my time, my money and my mistakes.

The rule is simple too, and I set it before starting: publish the numbers even if they're zero or negative, and finish the series however it turned out. If by the third week the business had died, the fourth part would tell how it died. Now that the challenge is closed, I'm publishing it in five parts, one a week.

Why do it this way? Because it's the only thing I can offer that anyone else can't. We don't sell data, we sell knowledge and experience, and experience includes the faceplants.

I'm not a business guru and I don't pretend to be. I'm just one more person giving it a go, with the advantage that I've been using these tools daily for a while and the disadvantage that this guarantees nothing.

Before choosing, I binged on promises. For two weeks in May 2026 I went through some thirty videos and a dozen articles and posts, without looking for anyone in particular to mock: I'm interested in the pattern, not the presenter.

Books made with artificial intelligence. What they promise: you publish ten books a month on Amazon and collect royalties forever. What they leave out: that thousands of people have had the same idea, that platforms set limits and that a book without readers is just a file.

Automation agencies. What they promise: you sell automations to companies for thousands of euros. What they leave out: that you have to find those companies, convince them and maintain what you build for them when it breaks.

Faceless channels. What they promise: generated videos that monetise themselves. What they leave out: the time it takes to meet the monetisation requirements and how little most generic content pays.

Online shops. What they promise: artificial intelligence picks your product, writes the listings and makes the ads. What they leave out: advertising, returns and margins.

Services "with AI". What they promise: you offer any service, add "with AI" to it and charge more. What they leave out: that the customer pays for the result, not the label.

There's a distinction that puts everything in order and that the videos usually blur.

One thing is making money using artificial intelligence: doing something faster or better that someone already pays for. A translator who delivers sooner, an accounting office that serves more clients, a designer who tries ten ideas instead of two.

A very different thing is making money selling artificial intelligence itself: courses, prompt packs, agencies. That's where the most promises are and where the most people compete offering the same thing. If everyone sells shovels in the gold rush, shovels stop being a business too.

So, can you really make money with AI? You can, but you have to subtract what almost nobody mentions: subscriptions, learning time, advertising and the hours spent checking what the assistant gets wrong. Income without subtracting all that isn't income, it's a headline.

My starting point: I could devote about 10 hours a week, mostly evenings and Saturdays, and a maximum budget of €500 for the whole challenge. With that in hand, I filtered with four criteria.

Initial investment. How much money you have to put in before selling anything. With my budget, anything that needs stock or lots of advertising gets complicated.

Available time. Ten hours a week is not much. If a business needs daily customer attention, it doesn't fit.

Competition. If thousands of people are doing exactly the same thing with the same tools, my advantage is zero.

What I know how to do. This one weighs the most. Artificial intelligence makes me faster at what I already know; in what I don't know, it makes me wrong faster.

I scored each option from 1 to 5 on each criterion (5 is the most favourable for me):

The shop and the channel go first: lots of competition, little for me to bring and, in the shop's case, money up front. The books have so much competition that even the low investment can't save them, and I knew that first-hand: I'd tried them in the winter (I tell the story here). The agency and the specific service end up tied, and the tiebreaker was time, which is exactly what I have least of. A warning about the table: the scoring is mine and it's subjective. Someone else, with a different trade and different hours, would come up with a different table, and that's fine.

I used the assistants as study partners, not oracles. A prompt, for beginners, is simply the message or instruction you write to the assistant: a question, a command or a text with directions.

I worked with Claude to organise and debate, and with DeepSeek and ChatGPT to cross-check. One of the real prompts: "I have 10 hours a week, €500 and experience building artificial intelligence systems for companies. I'm giving you five business ideas. For each one, tell me what would have to happen for it to earn me €1,000 net a month within six months and what's most likely to go wrong. Be harsh".

What it contributed: it organised my ideas, found gaps I hadn't seen and, when I asked, played devil's advocate with some flair. What failed: when I asked for data on the market for generated books on Amazon, one of the models confidently gave me an average income figure per author that I couldn't find in any source. Another stuck to generalities like "find your niche and add value", which is the equivalent of telling a castaway to look for land.

How I checked it: any figure it gave me I looked up in original sources, and whatever I couldn't find, I threw out. Artificial intelligence saved me hours of organising ideas, not of checking data.

After the filter, I went with a small, very specific service: building FAQ assistants for local shops, with a fixed price and monthly maintenance.

Why that one? Because it scores high on what matters most: it's something I already know how to do, the investment is almost entirely my time and there's hardly any competition focused on small shops in my area. Risks I could see from the start: that shops wouldn't want to pay a monthly fee, that each client would ask for something different and eat up my hours, and that maintenance would grow faster than income. What scared me most: selling. Building I can do; convincing someone to pay me, a lot less so.

The discarded finalist was the general automation agency. It was left out because I'd already tried it in the spring (I tell the story here) and it demands sales time I don't have with ten hours a week.

Before starting, I defined the goal, so I couldn't cheat myself later. For this challenge, a salary is €1,000 net a month, and the deadline, six months from launch.

How I measured it: income minus all expenses (tools, subscriptions, advertising, whatever), plus my hours logged. If I make €1,000 working 80 hours a week, it's not a business, it's a punishment.

What would count as failure: not reaching half the figure within the deadline, or reaching it at the cost of more hours than I have. I review this figure in parts 4 and 5.

What's coming, one part a week:

• Part 2: the launch, next week.
• Part 3: how I got the word out.
• Part 4: the numbers.
• Part 5: whether it was worth it.

You can follow the series and laugh with me, or at me, depending on how things went. Or better still: take your own idea, run it through the four criteria and make your own table. If something interesting comes out, I'd love to hear about it.