About AI agencies and why this whole ecosystem makes me somewhat distrustful

I've been watching how this whole "AI agencies" thing is developing, and honestly, it gives me a mix of frustration and disappointment. It's not something new (we've already seen it with cryptocurrencies, dropshipping, and practically every new technology that appears) but seeing the same pattern repeat over and over again is disheartening.

The pattern that always repeats

Lately I see many cases where someone claims to be billing, for example, €35,000 monthly with their AI agency. Curiously, that same person usually has a "community" or "academy" with about 500 members paying €79 per month. If you do the math, that's almost €40,000 just in training.

I'm not saying it's wrong to diversify income, but it makes me think: if the main business really worked as well as they say, would they need to dedicate so much time and energy to selling courses? It's something that doesn't sit right with me, honestly.

What they don't usually tell you about the numbers

One thing that especially bothers me is when people talk about revenue as if it were net profit. Anyone who has run a business knows there's a huge difference. A real agency has considerable expenses:

• Salaries if you have employees
• Office or coworking space rental
• Computer equipment that needs replacing
• Essential software licenses
• Accounting and consulting
• Taxes (I save on some of this by not living in Spain)

In the end, from those €35,000 in revenue, if you're lucky you're left with €8,000-12,000 clean. And that's if everything goes well and you don't have unexpected expenses.

But there's more, and anyone who has worked in agencies (whether software development or AI) knows this: there are good months and bad months. These people always show you their best month, but they don't tell you that in August they barely billed anything because clients were on vacation. They don't tell you that when you finish a big project, you can spend weeks or months looking for the next client, billing much less or nothing. It's the reality of the sector that you only know if you've been inside.

The GDPR issue that particularly worries me

Something that causes me considerable concern is how data protection is being handled. I see people happily automating processes that involve personal data, sending them to ChatGPT or Claude without thinking about the implications.

I've seen cases of automations to process CVs, manage client data (names, emails, phones, addresses), even medical information. Is consent being requested? Are there data processing agreements? What happens if there's a leak?

GDPR is no joke, and the fines can sink a business. But of course, talking about this isn't as attractive as promising easy money.

The maintenance headache

Another thing that frustrates me is that nobody talks about maintenance. AI models change constantly. A prompt that worked months ago might give mediocre results today. Maintaining these automations requires:

• Constantly reviewing outputs
• Updating prompts when models change
• Dealing with API changes
• Managing service outages and usage limits

It's continuous work that clients don't always understand or want to pay for. But without that maintenance, quality degrades quickly. I know this from personal experience having to close an AI transcription SaaS, due to costs and among other things having to modify prompts every few weeks and adapt to model and API changes.

More sensible alternatives that almost nobody mentions

I find it revealing that almost nobody talks about more professional alternatives. For example, tools like Ollama or vLLM allow you to run open source models on the client's own servers. They control their data, pay for their infrastructure, and there are no privacy issues.

That said, let's be honest: setting up AIs in local environments isn't cheap. Depending on the speed you need, the concurrent requests you have to handle, or the context window you require, you'll need more or less VRAM. We're talking about machines with powerful GPUs that can cost from €5,000 to €100,000 or more. But if you value privacy, not depending on foreign companies, and having total control over your infrastructure, I think it's the way to go for many companies.

Or take the case of n8n, which is very trendy. What many don't know (or don't say) is that their terms of use don't allow you to have a centralized installation for multiple clients. You must set up an independent server for each one. This isn't just for legal reasons, but also security: if something fails, it only affects one client, not all of them.

But of course, this multiplies costs and complexity. It's not as pretty as saying "set everything up on one server and charge 50 clients."

What you really need to work with AI

It especially bothers me when they say you don't need experience or knowledge. Working seriously with AI requires understanding:

• Basic systems architecture
• Computer security concepts
• Applicable legislation
• Project management
• And above all, the ability to understand what the client really needs and understand their business/sector

It's not rocket science, but it's also not something you learn in a weekend watching YouTube videos.

My personal reflection

After watching the umpteenth video of "earn X per month with AI," I needed to write this. I can't help but feel that we're wasting the potential of an incredible technology by turning it into the new "get rich quick" scheme.

The worst part is that I know perfectly well how this will end. In a few months, when the market is saturated and clients have had bad experiences with poorly made automations, AI is going to have a bad reputation. Like what happened with blockchain, which went from being a promising technology to being synonymous with scam for many people.

And meanwhile, those who really know the subject, who could do useful and well-made things, are going to have to deal with burned clients who don't trust because "I already got scammed once with this AI thing."

It's frustrating to see how the cycle repeats. New technology, impossible promises, saturation of snake oil salesmen, burned market, next technology. And start over.

If you're reading this and thinking about getting into the world of AI, do it, but do it right. Really learn, don't believe the siren songs of easy money, and above all, be honest with yourself and your clients about what you can really offer.

Because at the end of the day, those who really win aren't those who promise the moon, they're those who deliver real results, even if they're modest. And those don't need to sell courses at XX€ per month to make a living.

PS: If you really want to learn about AI, there are excellent free resources: official tool documentation, YouTube, academic papers... You don't need to pay €€€ per month to anyone to learn the basics.