Good Morning Reader it's Maria,

Am I being paranoid?

I've been using AI tools for close to four years now. ChatGPT was my first love, then I discovered Claude and, in a beat, I switched over. Yes, I know, not very loyal of me....

But these models are always improving, and I quickly realised that constantly challenging them is the way to get the best results. These last few months, though, I've found that Claude's quality has dropped, while ChatGPT has gotten a lot better. Which is weird, as each update is supposed to make them better at what they do, no?

So I had to ask myself the question: Am I making this up? Is it just me, or is something happening here?

If podcasts are more your speed, I've got you covered; there's a discussion about this topic available now here.

THE BIG IDEA

Turns out I'm not paranoid (phew!).

I did a little research, and I found that media outlets such as Fortune, VentureBeat and Inc have all written about Claude's quality drop. And to be fair, it's not only Claude, there are also articles about ChatGPT. It's like your phone; they're constantly getting upgraded, and every time there's an update, everything changes.

Here's my point.

I see people choose an LLM (large Language Model) such as Claude, ChatGPT or Gemini, get comfortable and then stick to it. And I get it. It's easier and cheaper to have it in one place, but I think it's a mistake, and here's why.

We've watched this exact story play out before with social media. Build your audience on a platform you don't own, and you're one algorithm change or one ban away from losing it all.

This isn't hypothetical; recently, a ChatGPT user, a paying subscriber since 2023, had his account suspended overnight without any warning or explanation. He lost his subscription, his custom GPTs, and all his work, and he's still fighting to get it back.

Google's own support forums tell the same story on Gemini: users locked out for weeks with appeals sitting unanswered; it's already happening across all platforms.

And what happens when the prices go up? They will, and this isn't me being dramatic Reader.

These tools are still incredibly cheap for what they do: twenty euros or so a month for something that used to need a small team; it's a no-brainer for most people. But that won't last forever.

None of these companies are making real money yet.

OpenAI is on track to lose something like $14 billion this year. Read that again. 14 billion. By some estimates, they're spending more than a dollar for every dollar that comes in the door. Anthropic's doing better; its losses are shrinking, and it's hoping to break even by 2027, but right now it's still spending way more than it's bringing in too.

And both of them are throwing eye-watering amounts of cash at computing power just to keep the lights on; OpenAI alone has committed over a trillion dollars to that for the years ahead.

That money isn't coming out of thin air. Someone has to pay for it; for the moment it's the investors; eventually, it will be us.

OK, no need to panic! Nobody's pulling the plug overnight, but I don't want you to be shocked when the prices start going up.

This is exactly their strategy to hook us: cheap or free, so you pour in your time and all your data. Once you've built months or years of your business inside one tool, walking away feels almost impossible, even if something better or cheaper comes along.

There's a name for that: sunk cost fallacy. The more you've already put into something, the harder it is to leave, even when leaving is the smarter move. That's the model, and they know it works.

And finally, here's the last reason you shouldn't put all your eggs in the same basket.

Twenty years managing teams, there's one thing I never did: hand a junior's work straight to a client without reading it first. It's an opportunity for you to catch mistakes and for this person to learn from them.

I treat AI exactly the same way. Think of it less like some all-knowing oracle and more like a ridiculously fast, incredibly well-read golden retriever energy-like intern on your team. Brilliant one minute, confidently wrong the next.

If you get too comfortable with one LLM, you become complacent; that's when you stop checking the outputs, which leads to AI slope.

None of this means you need to spend your life testing every new AI tool that launches. Please don't! It's overwhelming.

It just means: Stay a little sceptical, keep questioning the answers and keep an eye on what else is out there.

Be smart and don't build your entire business or way of working on top of ONE platform you don't control.

Don't know where to start?

​Book a free consultation and let's chat!

THE ACTION STEP

First, list every AI subscription you're paying for.

For each one, ask yourself: have I opened this in the last month, and am I paying because it's still the best tool for the job, or out of habit?

Second, make a backup. Here's exactly where to go:

• Claude: Settings > Privacy > Export data
• ChatGPT: Settings > Data controls > Export
• Gemini: no in-app option; you'll need Google Takeout (takeout.google.com). Skip the "Gemini" checkbox; that only grabs your custom Gems, not your chats. Go into My Activity and select Gemini Apps instead.

Each one emails you a download link. Some arrive in minutes, some take days so don't wait until you need it to find out which.

Have you just signed up? See all previous newsletters here.​

AI MADE SIMPLE

Since we're on the topic of keeping your own house in order, here's what I've been doing with mine this week.

My data was a mess, too many folders, so many resources, and it had basically grown out of control.

This is one of the first things I tell my clients when I audit their business: your output is going to be as good as the data you input and how easy it is for the AI to find information needed. That's why it's super important to have your files sorted.

So had to take a hard look in the mirror... the shoemaker's kids and all that.

I finally sat down and sorted it out with ChatGPT's help, and here's the process I'd actually recommend.

Start by getting a full-text list of your folder structure, so nothing gets missed or misread:

• On a Mac: select your main folder in Finder, right-click, hold Option, and choose "Copy as Pathname"
• On Windows: open your main folder in File Explorer, click into the address bar, type "cmd" and press Enter. In the window that opens, type "tree/F" and press Enter, then select all and copy

Paste your whole folder structure into ChatGPT or Claude in one go, not folder by folder, so it can see the full picture and give you one coherent system. Try this prompt:

"Reorganise this folder structure using the PARA method: Projects (active client work), Areas (ongoing responsibilities), Resources (reference material), Archive (done and dusted). Attached is my current structure. Suggest a cleaner structure with clear categories and subfolders, flag anything that sounds outdated or worth archiving, and give me a simple step by step plan I could put in place in under an hour."

If you have folders with client names or sensitive info on them, I'd suggest renaming them first or pasting the folder structure into Word, removing any of those, and then pasting it into the LLM. Use folder names only, in general terms.

There is an even simpler way. Claude Cowork or ChatGPT's Codex can now do this for you directly inside your files. Personally, I've heard too many horror stories, so I'm not ready to give an AI access to all my folders (it has access to selected ones). But if you're comfortable with it, it's an option.

That's all for today Reader

Have a great weekend!👋🏼

Take Care

Maria

PS: If you want to explore what working together looks like, book a free call

PPS: If you enjoy these emails and want to do something nice, you can buy me a coffee 😉

​​​