I don’t know about you. But I’m getting this really strong Déjà vu with LLMs (Large Language Models). As if I’ve been played before and I’m about to get sucked into this game again.
Not to be too dramatic or anything, but here is what happened.
Over the course of 10 days, I took a leap of faith and committed to the $200-per-month plan to get “50% of the time” access to Fable. Still not sure what that means.
All I know is that I watch the bar that says weekly limit, and when it hits 100% and goes red, I’m out — waiting for next Thursday, 5 a.m., for the reset.
If you can’t tell, the game already started.
Over 10 days, I watched my limit get consumed at varying degrees — something I’d gotten familiar with from using Opus, Sonnet, and their sidekick ChatGPT.
But before I got sold on this investment, I had encountered an animation bug in React Native that none of the other models could solve.
To my surprise, Fable fixed it on the first go — by searching online rather than endlessly trying to reproduce the bug, relying solely on a screenshot I gave it.
I was impressed. Mind you, my background is in software engineering, so I was double impressed.
Loading an animation in a browser and catching the exact millisecond it fails is still something LLMs are bad at. Because of race conditions and all for the tech-savvy.
A dilemma of sorts that kept happening, even though I had explicitly said to the agent file, "Don’t ever test animations."
You burn tokens like there’s no tomorrow and end up back where you started — frustrated, fighting an opinionated LLM with a big unresolved bug.
Don’t go there. It’s dark.
So the game continues — and you start feeling the resistance.
Now, in those 10 days I accomplished what would have taken me 6 months. So this is not a rant about how expensive LLMs are, nor a plea to stop using them.
But as always, there’s a catch.
Sharing my marvelous creation with a friend to test, she asked me how I came up with this — a fully fledged app which you will learn about in a minute — and instinctively I said:
"It was me and Claude."
And then a few messages later I followed with:
“And of course 15 years of product development lol minor detail 😂” — verbatim.
The trap was set, and I was falling into it headfirst. The temptation was so extreme that the roles had started to mutate.
Who is the engineer and who is the machine? Who is programming who?
And that’s where it hit me.
You see, not so long ago 4 years to be exact I launched my first book, Reality Check.
The day I did that, I forcefully and begrudgingly entered the social media game to spread the word about it.
So to the whims of the machine learning systems behind these algorithms — the original LLMs, if I may say — I had to calibrate my content.
And I hated it.
One day I reached 1,000 people; the next, 500. An overnight loss I had to readjust to so my reach didn't falter. And at some point I gave up. Well, frankly, at multiple points.
The creator was at the mercy of the platform they were creating on — rather than the message their content had to deliver.
And there. Right there.
Déjà vu.
That’s what is happening with LLMs, and it’s deeper than the role reversal.
Over those 10 days, and another 10 that followed, I started noticing the patterns.
One: if you're spending more than $200 per month on your LLM, the roles have reversed — you are being played. You are now optimizing for the LLM, the same way you once optimized your content for whatever algorithm was ranking and placing it that month.
Two: context is both a blessing and a curse. The more you build, the more your agents.md file grows — and with it, the constraints on what you can do. You get stuck in a rabbit hole where half your context window is loaded before you even start. Split it into agents. Trim it ruthlessly. Most of it is useless. Same as your hashtags.
Three: the opinionated algorithm — your unapologetic LLM. This is where it truly becomes a game of social media. The model has tastes, same as the algorithm did. Fighting them burns tokens, so you give in. And slowly it’s shaping how you build, the way the algorithm shaped what you made.
Yet all is for naught. As for better or worse, here it is.
That app my friend tested is called Exercise Bites — a two-minute movement practice built around questions, because I needed it before anyone else. It’s now in public beta: exercisebites.fit
Try it. I’ll respond to every single one of you — happily.
Before Thursday, 5 a.m., of course.
Thus, the dependency is born the same day social media told you consistency is key and to post every day. Can you see it?
If not yet, that's okay. Don’t stop playing. But do remember: you know what good looks like — and when you don’t, it lives with your users’ feedback or the experts in their respective fields.
Nothing new here; the old playbook is still the same.
And after all of this, I don’t know if I nailed it — the marketing or the development. All I know is that the list of algorithms I keep contending with just keeps growing.
And so does the cost. $200 or $8000 is only the part you can see. The real cost is the behavioral one shaping you or your organization in every corner.
So, as the lines between you and your LLM blur, a fair warning — not a judgment call.
And even then, keep playing.
Hope this helps you, and best of luck on your journey!
Carlo Mahfouz