We cant stop thisā¦ we need to flow with it, peronsally I am not worried, I have a solid experience at this point about how it works and seeing how messy it is and the only way I can use it in my work is modular and only if you have solid skills the idea of doing somethign without knowing anythign about that somethign simply dose not work and it will never work!
1 Like
Yes, we canāt stop it, it is what it is. I am not worried, I have a healthy subscriber base and all is good but from a business perspective I am just thinking about the ways in which to get in front of it and be prepared.
So far the conclusion Iāve reached is that itās about Brand and Media.
At the end of the day, the customer doesnāt know anything about the code and the code doesnāt really matter. Itās all about who is first in Google and who the AI recommends.
And you canāt vibe your way to a top brand.
So Iām investing heavily in advertising and articles these daysā¦
You are in a very good place if you ask me!
1 Like
Ok guys I used today AI for about 40 min with Codex 5.3 it charged 50$ this is insaneā¦ which model do you use, with this prices is impossible to use it production wiseā¦ I guess is back to old way programing from no on 
And I only used it in modular steps. If you use vibe coding to build an entire app now, it can cost you thousands of dollars
I go mainly with ChatGPT Plus + Cursor
Cursor is doing a very good job and it costs $20 / mo - and I havenāt hit this planās limits yetā¦
Thanks to Stefan for mentioning Cursor to me.
2 Likes
Last month is was reasonable but now codex 5.3 that is what I uised is insane literally each prompt is taking 1-2 $ā¦
Thinking of how they played this game man so cheasy they l;et us think that AI is really cheap infact is more expensive than a human and is such a s*it showā¦ every time I use it I get angry but I hve to say it saves time by a lot and in some cases but overall is a mess, I really donāt see how this could replace anything at this point, you can see froma mile away that it dose not have any logic but works more like a drunk really good engeer, sometime it makes you think is ok not point to code anymore and other times yourealize that without skill AI is crappy and the good thing for this AI companies the closer you get to the end of your project the more you use it and spend more tokens.
Personally I donāt see how this techonology will survive short therm, it is to buggy and too expensive!
About Cursor, I believe itās only a matter of weeks before it aligns with the rest of the market. Otherwise, they risk going bankrupt. These GPUs are expensiveāreally expensive. They become outdated within 2ā3 years and consume a massive amount of power. The whole thing feels like an illusion thatās starting to fade.
Everywhere I look, I see unhappy people for all kinds of reasons, and the younger generation seems to be increasingly skeptical of AI. Part of the blame lies with CEOs and industry leaders who spent years promoting apocalyptic scenarios in their speeches. Recently, Iāve noticed a shift in the narrative: suddenly itās no longer about AI replacing us, but about AI creating more jobs. Now itās being presented as a positive force again. At this point, though, many people simply donāt trust them anymore.
The way I see it, this bubble is eventually going to burst. Very few people can afford the prices of the top-tier models, and many of the underlying issues remain unresolved. The entire industry feels rushed, and AI was marketed as something far more capable than it actually is. The expectations that were created are nowhere close to reality.
1 Like
I use only free ChatGPT and Gemini for parts of code and for text writing.
What exactly more benefits are available with paid AI if you use it only for development?
1 Like
If you havenāt tried it, give Codex a try as well (or is that what you mean by ChatGPT Plus)?
It seems to have 2 great strengths:
- Itās very methodical, it really listens and is careful
- Itās actually quite good at UI design and subtle things and effects
That said I havenāt worked with Cursor in a while myself so I no longer know how it compares.
Claude Code seems to have occasional strokes of genius but is usually a lazy bastard (doesnāt listen, doesnāt think things through etc) so Iām not a big fan.
1 Like
The agents are really good you would be surprised to see the difference!
1 Like
I donāt think Iāve tried any agent besides Manus.
What do you recommend, what should I try? What are they great at?
GPPT 5.3 codex is a beast literally, the best one is Open Claw but that one is 50x more expensive in a single prompt you can spend 50$ easy!
With giudance you can really push things with this top agents, but it got really expensive, using it an entire month heavy load it would cost you 2000$ - 3000$+ the only way to use it now is on really difficult tasks other than that we need to return to the old way of codingā¦ in the end nothing really changed the AI ilusion is calibrating really hardā¦ and the buble will pop hard becuase nobody can pay this prices for AI is impossible to make it workā¦
Of course in the beggining when you could use 100 agents to vibe code some app it was ok for this vibe codersā¦ now it will cost you too much, the profession of vibe coding died really fast
1 Like
I expect codex 5.3 agent to be free to use locally in about six months or max one year by then we will also h ave some sort of hardware to run it so we will use it freely, also devs all over the world are experimenting with this stuff there is no way it will be managed just by a few companies, things will chage and for the better for us normal people!
1 Like
Something bad is happening in the AI industryā¦
After the AI companies offered their products cheaply and as companies started to adopt them these supplier companies āfound outā that they donāt make money. The point is, they āfound outāā¦they knew well in advance what to expect.
But, like any self-respecting drug dealer, they sold a cheap ādrugā and then restructured and increased the prices! And as companies got a taste of this ādrugā, they became addicted. And when youāre addicted, you pay whatever you want. Thatās how some consumer companies were surprised by their paychecksā¦some found out that they were paying more for AI than they would for the people they fired. Without the productivity gains being reflected in the companiesā profits.
Microsoft, which is also an investor in OpenAi (the maker of ChatGPT) and Anthropic (the maker of Claude), has told its engineers in a large division to stop using AI in their work. Specifically, to stop using Claude because the paycheck has become too high. 30% of the code written by Microsoft engineers is written with AI Now they will move to a tool that they own and that is cheaper to use.
Uber used up its AI budget for a year in 4 months. 70% of the code written is written with AI.
So employees went into debt, using AI for, you know, everything.
Amazon itself has reduced the requirements for all its engineers to use AI (hell, some engineers used AI to see what the weather was like! That was to show that they were using AI).
Just think about itā¦if a company has a product made with AI and that can have hundreds of thousands of lines of codeā¦when you want to make a change to that product, no matter how small, using AI then that AI has to review all those hundreds of thousands of lines of code (and that costs in tokens - because thatās the measure now) to figure out where to make the change. An engineer knows roughly where he should look and would go straight there without reviewing everything.
The bill for AI is not calculated the way it is calculated for a traditional software subscription. Do you have Microsoft Office? You pay that much for the subscription per month. It doesnāt matter if you use it 24 hours a day or not at all. Whereas with A.I. you pay about the same as you pay for electricity. Every time you turn on the light bulb or the radiator, the meter goes!
In AI the unit of measurement is not the watt or kilowatt but the token. A token can be a word, part of a word, a punctuation mark, a blank space, a number. For the English language, this would translate to about 100 tokens being equivalent to 75 words, more or less. And all our interaction with the AI is done through words (and when you interact with voice, it is also done through words - the AI translating spoken words into written words). And since many times what you ask the AI you donāt get back as you would like, then you have to repeat the procedures by changing the words. That is, you basically waste tokens until you get what you want. At least in image and video generation (which I use more often) many regenerations are made until you get what you want. And yes, images are tokens (numbers) too.
Companies providing AI have subsidized the cost of tokens since they came on the market (through investors who have put a lot of money into them)ā¦but in the end they also have to make money. And they can only do that if prices increase. And no, building more datacenters (i.e. even greater scaling of token processing) will not make them cheaper. Because the initial investments are extremely large (hundreds of billions of dollars) and must be recovered somehow.
Thatās why some companies will go public, to get money from the suckers, so to speak. Because the smart ones put money into these companies when they were just starting out. Now they need to be paid. Thatās what SpaceX will do when it goes public this month. Thatās what Anthropic and OpenAI will do too.
The drug of the last decade was social media (it still is)ā¦ the drug of this decade and the next will most likely be AIā¦
Remember when I told you a long time ago that nothing had really changed when it comes to our profession? If you donāt understand your craft, AI is just an expensive illusion.
In my case, I can only use it in a modular way for specific tasks. An entire app built with AI is a doomed project. Itās not scalable, itās difficult to maintain, and updating it becomes a nightmare.
You also missed one huge point: every time you want to update your vibe-coded app, the new AI agent has no real understanding of what your project is about. It will inject code that might fix one bug or add one feature, but at the same time it can break other things you donāt even know exist. Thatās the fundamental problem.
There is no future in blindly relying on this stuff. The future is humans and AI working together, but the human has to be genuinely skilled at what they do. Otherwise, it simply doesnāt work.
All these companies that bet everything on AI are getting a reality check. These models donāt think; theyāre sophisticated pattern generators. And theyāre becoming ridiculously expensive. In my case, I used AI for about two hours on a few modular tasks and got charged around $50. Not long ago, that would have cost maybe $2.
Yes, the AI companies played their cards well, but reality is hitting hard. To be honest, the technology becomes almost unusable when the costs start adding up.
The only scenario where I see it being truly useful is helping you create an applicationās architecture and assisting with complex tasks here and there when you need a faster solution. But if you let it take over the entire development process, youāll end up paying thousands of dollars and still have a project thatās essentially lost.
Even if you somehow get the app working, how do you know everything is functioning correctly if you donāt understand how it was built? Then, a few months later, when you need a bug fix, another AI agent comes in and injects its own sloppy code, making the application even more unreadable. Eventually, the codebase becomes such a mess that nothing works properly anymore, and no developer wants to touch it because reading through that chaos is pure insanity.
Long story short, things are starting to recalibrate. But look at the damage this hype has done to the industry: job losses, collapsing incomes, and engineers struggling everywhere. For me personally, income has dropped to almost nothing, and I know many other engineers in the same situation.
And donāt even get me started on these CEOs. Some of them are the worst of the worst. I donāt even want to talk about the Envato CEO because I always end up angry. That guy somehow managed to break something that seemed impossible to break. In a strange way, thatās almost impressive so youa re amazing mr. f*king Hichame Assi
When Envato eventually shuts down completely, and I believe it will, because Elements garbage and the marketplaces are already dea, Iāll at least have the satisfaction of seeing this idiot fail so spectacularly.
He inherited a treasure and somehow managed to turn it into garbage with all of his āhigh intelligence.ā
Ugh, now Iām getting angry again. Iād better stop, I wil lmake a video at some point about everythign that wente wrong after he came into power and how he destoryed things, all he needed to do is to listen to its authors, nothing else!
1 Like
AI is becoming unusable because itās simply too expensive. The only practical way to use it is for tasks that are genuinely difficult to code manually, such as shaders or other complex algorithms. Otherwise, your monthly bill can easily reach $1,000ā$3,000 for agentic coding alone, nd thatās if you know what youāre doing. I am reffering to the top models like codex gpt 5+ and others who are more expensiveā¦ older models are cheaper but still not even close to what we were used to!
For vibe coding, the cost can grow exponentially. You might end up building an app and receiving a bill of $10,000 or more. This is the real cost of AI.
These gready evil bilionairs CEOās would burn the world for profit. The bubble is popping, guys, the future that they want for us is starting to crumble the masive money investment is done from now one is reality check!
AIās been a huge help for me with both Dev and Support and I donāt think I spend more than $100-200 per month.
I think some people are also clueless or have some kind of āhustleā mindset where they think having 10 agents working overnight will achieve anything meaningful.
If you use AI for clear, well defined tasks without huge context window, it seems pretty cheap.
I could pay 5-10x what Iām paying now and it would probably still be worth it for me.
I am glad it workd for you, for me is too expensive at this point ā¦