---
title: "When Should You Use The Magic Thinking Rock"
url: "https://bravr.ai/blog/when-should-you-use-the-magic-thinking-rock"
---

# When Should You Use The Magic Thinking Rock

## Rock Whispering

By vuk 5 October 2026 Rock Whispering

**Key Takeaways**

*   AI is a multiplier, not a replacement. Anything times zero is still zero.
*   Use the rock for your strengths, so you can tell when it's wrong.
*   Boring, repetitive, checkable tasks are prime rock territory.
*   High-stakes work you can't verify needs a human in the loop.
*   If the rock can't get the context, it can't do the job.
*   One shareable decision matrix to rule them all.

Welcome back to the second entry of the Rock Whisperer guide! A guide where we teach you the ins and outs of the tech bros’ newest toy and Wall Street’s newest market crash! Since we’ve covered how to use the Magic Thinking Rock in last week’s edition, which you can browse [here](https://bravr.ai/blog/how-to-get-the-magic-thinking-rock-to-do-what-you-want-it-to/) (well, I’m using “week” as a figure of speech, it was actually three months ago, but oh well), this week’s edition focuses on the “when should you use the Magic Thinking Rock.”

## The Backstory

A couple of months ago, I decided I was going to become a techno producer. Not because I have any musical talent, mind you. My entire musical career consists of two years of mandatory recorder lessons in primary school and one very confident karaoke performance that I am contractually not allowed to talk about.

But I had the magic thinking rock. I had just finished writing an entire blog post on how to get it to do what you want. I was, by my own estimation, a certified rock whisperer. How hard could a techno song be? It’s basically the same four beats for seven minutes, right? (Techno people, please don’t email me. I know. I know now.)

So I fired up one of those AI music generators and typed in my masterpiece of a prompt: “dark, driving warehouse techno, 130 BPM, hypnotic”. Hit generate. Leaned back. Started mentally drafting my Boiler Room set list.

What came out sounded like the hold music of a dentist’s office that just discovered dubstep in 2011.

No problem, I thought. I’m an engineer. Engineers iterate. “Make it darker.” It got… louder? “Make the kick punchier.” It changed the kick. Was it punchier? Reader, I have no idea. I didn’t know what punchy was supposed to sound like. I knew it was wrong, I just didn’t know _what_ was wrong. Same way you know a dish is off but can’t tell if it needs salt or an entirely new chef.

Three hours and about forty generations later, I had forty different flavours of bad, and zero ways of telling which one was the least bad.

Then I showed the whole sorry thing to a friend who actually produces music. He took the same tool, on the same account, and typed in things like “rolling offbeat bassline, open hi-hats on the off-beat, sidechained pads, long reverb tail into the breakdown, build into the drop at 2:30”. Twenty minutes later, he had a track that I would unironically dance to. Badly, but still.

Same rock. Same tool. Same monthly subscription. Completely different results.

And here’s the kicker (pun absolutely intended): I can’t create a song with AI, because I can’t create a song without AI.

## Why This Happens

In the last entry, we talked about how the rock is mostly trained on beginner content. Here’s the other half of that story.

AI is a multiplier. It takes whatever you bring to the table and multiplies it. If you’re a 9/10 developer, the rock turns you into a 9/10 developer who ships in an afternoon what used to take a week. If you’re a 0/10 techno producer, well. Anything times zero is still zero. I checked. With a calculator. A deterministic one.

The problem is twofold. First, you don’t know the words. Every field has its own vocabulary, and that vocabulary is the steering wheel. My friend said “sidechain”. I said “make it darker”. One of us was driving the rock, the other one was yelling at it from the back seat.

Second, and way more dangerous, you can’t tell when it’s wrong. When the rock writes me a function, I can read it and see that it’ll blow up the moment someone types an emoji into a form field. When the rock writes me a techno song, I genuinely cannot tell the difference between “absolute banger” and “crime against eardrums”. And the rock, bless its silicon heart, will always tell you it’s a banger.

That’s the beautiful, terrifying thing about the rock. It is always, _always_ confident. It’s confidently right and confidently wrong in the exact same tone of voice. The only thing that tells those two apart is you.

## What This Article Aims to Do

Same as last time, this isn’t a “10 AI USE CASES THAT WILL MAKE YOU A MILLIONAIRE BY FRIDAY” post. The fountain of youth people are still paying me, so that myth stays busted, and the millionaire-by-Friday people haven’t called yet.

What I want to give you is a framework. A handful of questions to ask yourself before you open up a new chat, so you know whether you’re about to save yourself three hours, or spend three hours generating forty flavours of dentist office dubstep.

At the end, there’s a decision matrix. Screenshot it, print it, tape it to your monitor, send it to your boss. Speaking of my boss, that raise. Still waiting. Share the post, you know the drill, pls and thank you.

## #1 – Know thyself (and thy rock)

The first and most important rule: use the rock for your strengths, not to cover your weaknesses.

I know, this sounds backwards. Surely the whole point of AI is to do the things you can’t? That’s what the ads tell you. “Never written a line of code? Build an app in minutes!” And sure, you can build an app in minutes. You just won’t notice that it stores everyone’s passwords in plain text, and you’ll find out the fun way. (Allegedly, again.)

The rule of thumb I use: if I can’t judge the output, I can’t use the output. I can judge code. I can judge (most) technical writing. I can judge whether an email to my boss sounds like me, or like a LinkedIn influencer having a breakthrough at 5 AM after an ice bath. I cannot judge techno, legal contracts, or whether that mole looks suspicious.

Which brings me to a confession. Remember how in the last post I said I couldn’t get AI to write content to save my life? This post I wrote together with the rock. Not because I suddenly got good at prompting (I did, a little), but because writing is something I actually know how to do. I wrote the intro, I wrote the structure, I knew exactly what I wanted it to sound like, and whenever the rock got it wrong, I could tell. That’s the whole trick.

Now, this doesn’t mean you can’t use AI to learn something new. Quite the opposite, it makes a great tutor, more on that in the matrix. It just means you shouldn’t ship something you can’t evaluate.

## #2 – The intern test

Here’s a mental model that works surprisingly well: treat the rock like a very fast, very eager, very well-read intern who has never once in their life said “I don’t know”.

Before you hand a task over to the rock, ask yourself two questions.

1.  **Can I check the work?** Is there a quick, reliable way to tell whether the output is right? Code either runs or it doesn’t (mostly). A summary of a document can be spot-checked against the document. “Which stock should I buy?” cannot be checked until it’s too late.
2.  **What happens if it’s wrong?** A typo in an internal Slack message: nothing. A typo in a contract: well. A made-up court case in a legal filing: a real lawyer actually did this, the judge was not amused, and it made the news around the globe. Google it, I’ll wait.

If the work is easy to check and cheap to get wrong, let the rock rip. If it’s hard to check and expensive to get wrong, ask yourself whether you’d let an intern send it out unsupervised. You wouldn’t. So why would you let the rock?

## #3 – Repetition is the rock’s love language

Remember skills and harnesses from part one? This is where they earn their rent.

The best tasks for the rock are the boring ones. The ones you’ve done a hundred times, will do a hundred more, and could do half asleep. Writing meta descriptions. Formatting reports. Turning meeting notes into action items. Writing the same “per my last email” email for the fifteenth time this month, with significantly more polite wording than you’d personally use.

My rule: do it once by hand. If you’ll do it again, write it down as a skill. If you do it every single day, put it behind a harness and forget it exists.

The reason this works ties right back to #1. If you’ve done something a hundred times, you know exactly what good looks like. You’ve got the vocabulary, you’ve got the taste, and you can spot a bad output from across the room. The rock does the typing, you do the judging. Division of labour. Adam Smith would be proud.

The opposite is also true. If it’s a one-off thing that needs a ton of nuance, explaining it to the rock might take longer than just doing it yourself. Not every nail needs a magic hammer.

## #4 – Does the rock have its lunchbox?

Last time, we compared the rock to a horse with blinders. It only sees what you put right in front of it.

So before you ask it to do something, ask yourself: can I actually give it everything it needs to know?

Some context is easy to hand over. Here’s the document, here’s the spreadsheet, here are our brand guidelines that nobody has opened since 2019. Pack the lunchbox, send the rock off to school, Bob’s your uncle.

Some context is a lot harder. The fact that Marko from accounting has been sensitive about deadlines ever since The Q3 Incident. The fact that the client said “we love it” but actually meant “we’ll be going with another agency”. The vibe in the room. The rock was not in the room. The rock was never in the room.

If the task runs on context you can easily pack, great, pack it and go. If it’s context you’ll need again and again, that’s what RAG, fine-tuning and giving the rock access to your inbox are for (see part one). And if it’s the kind of context that only lives in your head and in hallway conversations, the rock will give you the soulless, cold email. Every. Single. Time.

## #5 – When to leave the rock in the garden

And then there’s the stuff the rock just shouldn’t be doing. Not because it can’t, but because it shouldn’t.

**Things where taste is the whole point.** If you can’t tell good from bad, neither can the rock. See: techno, my.

**Things where someone needs to be accountable.** The rock can’t be fired, can’t be sued, and can’t sit across the table from an angry client and say “that one’s on me”. You can. And you will.

**Anything medical, legal or financial where a wrong answer has real consequences.** Use the rock to prepare your questions for the doctor, not to replace the doctor.

**Things where the whole point is that _you_ did it.** Wedding vows. Your mom’s birthday card. The apology text to your partner. If they find out the rock wrote it (and believe me, they will, because “I hope this message finds you well” is not how you talk to the person you live with), you’re in deeper trouble than when you started.

## The Decision Matrix

Okay, the bit you actually scrolled down for. I see you. It comes in two parts.

#### Part one: Should I use the rock at all?

Two axes. How well do you know the task? And how bad is it if the rock gets it wrong?

Dimension

Low stakes

High stakes

You know the task well

Let it rip. Check the output and ship it.

Rock as intern. The rock drafts, you review, nothing goes out unchecked.

You don't know the task

Rock as tutor. Use it to learn, not to ship.

Leave it in the garden. Bring in a human expert, and use the rock only to prepare your questions.

My techno song? Didn’t know it, low stakes. Rock as tutor. Which is fine! I could’ve used it to learn what a sidechain is. Instead, I tried to ship it to the group chat. Mistakes were made.

#### Part two: Which part of the rock toolkit do I reach for?

So you’ve landed in a quadrant where the rock is allowed to play. Now pick the right tool from part one.

If the task is…

Reach for…

From part one

A one-off, simple job

A well-written prompt

#1 Prompting

Specialised (code, maths, images, music)

The rock that’s best at that job

#2 Rock picking 101

Something you’ll do again and again

A skill, inside a harness

#3 Swiss Army knife

Dependent on your company’s knowledge or voice

RAG, fine-tuning, or access to your data

#4 Feeding the rock

Needed at 3 AM, or from your phone

An agent gateway

#5 Discord rock

#### The 30-second gut check

Not a fan of tables? Ask yourself these before opening a new chat:

1.  Could I do this myself, given enough time?
2.  Can I tell if the output is good or bad?
3.  What’s the worst thing that happens if it’s wrong?
4.  Am I going to do this more than once?
5.  Can I give the rock everything it needs to know?
6.  Does it matter that I’m the one who did it?

If you answered yes, yes, “nothing much”, yes, yes, no, congratulations, you’ve found yourself a rock-shaped job. Anything else, and it’s back to the matrix.

## Conclusion

Look, the rock is incredible. I use it every day, the team at Bravr uses it every day, and I’d hate to go back to a world without it. But it’s a multiplier, not a magician. It makes good people faster, and confused people confidently wrong.

So use it for what you’re already good at. Let it handle the boring, the repetitive and the checkable. Keep the stuff that needs taste, accountability or an actual human heart for yourself. And if you want to make a techno song, maybe learn what a sidechain is first. I’m told it helps.

As a wise man (me, three months ago) once said, Mind Reading Magic Thinking Rocks are billed separately. Turns out, so are Techno Producing Magic Thinking Rocks.

Next time on the Rock Whisperer guide: \[who / where / what, TBD\]. Until then, the raise. Share the post. My boss reads the analytics. Pls and thank you.

## Interested in any of the above?

Not sure where the rock fits in your business? We use everything described here in our day-to-day to figure out exactly that for our clients and solution partners.

[Get in touch](/contact/)

Thanks for coming to my TED talk. Again.

![Marinković Vuk](https://cdn.bravr.ai/wp-content/uploads/2026/06/cropped-1778500484353.png)

#### Marinković Vuk

[](https://www.linkedin.com/in/marvuk/)

Software Engineer & Technical Project Manager

AI & ML · Development · AI Infrastructure

Vuk Marinković builds AI products and the systems that power them. At Bravr, he focuses on turning ambitious ideas into practical tools, with an emphasis on AI agents, automation, and real-world product execution. He enjoys solving difficult technical challenges, moving quickly from concept to implementation, and finding ways to make complex technology feel simple and genuinely useful.

Vuk enjoys exploring where technology is heading next, usually by building things before there’s a clear playbook for how they should work.

[Back to Blog](/blog/)