After reading all of the grievances with the new pricing models and all the unhappy people on the forum and all over the internet. I sat back, took a sip of my coffee and was transported back to early March 2000.
I read a post someone else posted and then it dawned on me. We are 100% heading to the next big internet bubble burst.
None of these Ai model companies are making back what they are losing. Anthropic, Google, OpenAI, are all suffering massive losses on their models and are running deficits beyond comprehension, I mean in the Billions of billions.
Itâs just a matter of time where this will just not be sustainable anymore, and investors are going to pull out, much like a drunken night with Tatianna after realising you donât have money to pay for the service.
So here is my future predication. I genuinely believe, unless some magical âthingâ comes along that itâs just a matter of time before the legs buckle under the camel called the Ai Bubble. Ai wonât go away, but it will be rebranded and just like the dot-com bubble, of early 2000 will have to be approached in a totally different way.
Salute my fellow vibers!
P.S I guess the companies need to go and start rehiring all the programmers they fired
Back to the stack overflow era :)))))))))))))))
But itâs hard bro. Even if AI collapses, it will still continue to exist in some way. Because it is the ultimate product of humanity, it must exist.
Ai seems to be here to stay. With GPT-3.5 it may have looked like its not good enough for professional usage but one thing that you may not see is how successful lots of developers are using AI and how much time it saves.
Sure some adjustments are happening (cost and billing structures) across AI providers and it may limit access to those with less budget.
However overall cost seem to be moving towards customers, for example a developer has 10 clients and works for some of them on development with AI. The developer can include the AI cost into their cost structure and pass the expense down to client.
While there may eventually be a correction in pricing and so far the industry is making significant improvements across all verticals.
I hear what you guys are saying, but you cannot tell me it doesnât look good. Sure GPT 3.5 is awesome for engineers, but most people are not nearly on that level. Once the heavy, high models drop (Claude, GPT5 etc) the subscriptions will also drop. Most people on here complaining are âvibersâ and want to generate large amounts of code, that works, regardless of, if its slop. The better models do that very effectively. Engineers will not be able to keep the engine running, as they do not really need AI in the first place.
These companies are in a race, they just donât know where theyâre heading. Weâll see where it settles, but there is no doubt AI is here to stay, but maybe not how it currently looks. I could see it also getting super expensive and basically paying $16/mo for something like Cursor will be nonexistent in a year once the value is understood. Definitely some jobs will be gone forever. Programming has always been a bit inhuman, so probably not a terrible thing that a significant amount of it will be handled by AI in the future. Maybe a new programming language or way for humans to interact with code will be developed since a lot of the syntax and other nuances can be handled by AI in almost all situations. Things are in a state of flux and tbh I have no idea where things will end up. We may all be learning skills that will be completely useless in a few years. So in a way, not only are the companies racing to an unknown destination, so are all of us trying to keep up in the wake.
I believe LLMs will stay, they are not going anywhere. But they will become more useful locally. To use the level of inference Claude offers will become super expensive. Which in the end will hurt all the people who jumped the train of building software without knowing how to do it. There will be some âno codeâ style tools, probably hosting their own models and charging based on it, but price of using this through tools like Cursor is gonna become such that the price will replace developerâs salary, and I am not aware which company will pay for such a tech debt generator (there will be some for sure, question is only if enough to make that model profitable).
While I donât think the tech will vanish, there will definitely be a stock valuation collapse at some point when it starts to set it that LLMs in particular cannot do everything they were promised to do in a way comparable to humans. They will become tools to increase the productivity of competent humans, not replace them. Eventually people will figure out what to use that additional productivity for. But weâre all still going to have to know how to code at the end of the day to make anything that can last long term.
Iâm already seeing projects people created using vibe coding showing how they are unsustainable because AI just doesnât write maintainable code. Not to the degree needed for a project to survive long term.
AI will never generate code superior or equal to a truly competent programmer. There is value in the craftsmanship of code that goes beyond its functionality. I deal with a platform that was originally written by humans without any long term plans and its stalled a multi-billion dollar companyâs ability to improve on its existing feature in any truly meaningful way for several years now because the code was not written with any vision. AI has no vision. It can increase productivity of competent programmers using it to assist them in writing good code, or produce dead end technical debt. There is no way to get good code from an AI without knowing how to code in the first place.
We are early on. Most the tools out right now are on the frontier. The evidence is in progress, if things keep progressing the way they are, tools will become better and AI more competent. I agree they canât compete with true experience, but there are only so many experienced people who can only work so many hours a day for a competitive price. There is a balance, but I think itâs too early to judge the true extent AI can program. Simply humans are programming in a computer language. It makes sense itâd be more efficient to have the computer program in its own computer language. The fact AI can gain context of 100k lines of code in a matter of seconds already makes it a superior problem solver in regards to programming⌠at some point it will surpass most humans. This is like saying humans can solve arithmetic faster than a computer. Also we have to imagine the best tools are not public and accessible by the plebs. Certainly not for $20/mo or even $1000/mo.
Genuine innovation and vision like you said are really the only valuable skill that someone in tech can posses. And most people are not that creative so AI probably has them beat there too.
I am also far from convinced LLMs in any way mimic what the human mind does. Its based on the assumption we think in terms of our language and what comes next from that. I know personally its far from that - though I do not know if my personal thinking style is what others experience or not as I have never been anyone else except in a school play. Language is so slow and cumbersome.
Iâve been wondering the same thing. I think the usage curve went fully parabolic faster than expected. Crazy thing is, according to reports, most usage is still more of the novelty usage: Ooh, this is cool! That is shifting, as more and more businesses leverage it for ever more tasks, but it still seems like the novelty ââ â â â â â â â around with AIâ is the main use case.
The power requirements, the cost of building new datacenters, the impact those datacenters have on the communities they invade (yes, invadeâŚits not a good scene where data centers move in and consume nearly ALL the power, water, etc.), is all negative for AI right now.
Makes me wonder if that âmagical thingâ you are abstractly referring to, is quantum computing. LLMs are all vectors, matrices and vector/matrix math and operations. PERFECT for the limited range of capabilities of quantum processors. I do wonder if the technology that saves LLMs will be broad scale quantum computing centers. For a given amount of power (and thus fundamental cost), a quantum computer could do FAR more than a conventional computer/cpu. There will still be that need for conventional, as there are things quantum computers cannot do. So quantum couldnât completely eliminate the need for additional computing power and processing needs. But by offloading the majority of the core model computation to a vector-processing beast that could be billions or trillions of times faster than a conventional cpu, that might be the MAGIC we needâŚ
Look here â heâs talking exactly about this, including Cursor, and he explains some really interesting points.
In fact, from what he says, it seems that the only ones likely to survive are Cursor, while other companies are basically just gambling until they run out of money.
He doesnât mention in the video that Cursor has already solved the problem, but itâs still fascinating.
I also didnât know that Grok is the most expensive to use via API â even more than Opus.
Itâs nice to imagine, but weâre still very far from having such a computer.
Much more likely in the meantime is that theyâll find ways to reduce the cost of inference.
Some really interesting viewpoints here, and I think most of us feel the same. I donât care what people say but for me personally, as a mid tier developer, AI has elevated my programming experience to a level that I would have never been able to do alone. If you watch the models and take time in your prompts you do not have to produce slop.
Donât get me wrong, my original post was not about bashing AI and hoping it fails because I am this purist. Cursor especially has made programming a 1000 times more fun, and I also think that is why so many people are so overtly passionate and also passive aggressive when they think the tools are failing. You basically donât want to live without it anymore.
But can I say something really mean. My hopes is that the âsiftingâ would push some folks away with no passion for code to just regurgitate the most horrible slop to try and push out their billion dollar ideas.
ANDâŚ
Its the â â â â engineers thatâs clogging the pipelines. The ones that â â â â â â â â about AI tools the most are the ones thatâs just abusing every instance of trying to squeeze out as much productivity from the models as they can. Can someone say âhypocrite.â
With all said and done, with all the shyte thatâs going on and the unhappiness about the prices etc, Cursor has changed the way I program and for most of us, for ever. Your tool is truly a game changer regardless of all the issues its facing now.
Thanks for the conversation. Nice to hear everyone feels about the same.
I will say the ones that complain the most about the pricing, hitting limits, etc. are the ones that were before AI already working to crank out a bunch of code for far less money than the expected end product required. AI has just doubled the expectation for output and cut the amount bid for it even more. Its not sustainable economics - because the world needs less slop not more.
LOL if you think its writing better quality code and better architected systems than an experienced conscientious developer would, you have no clue what youâre doing in this industry.
I have written a whole small town tourism application, that consists of thousands of lines of code, has analytics, user authentication with multiple roles, reviewing, clients can respond to reviews, favourites a complete and functional search, Client Dashboard, with Advance Analytics, Adding multiple business categories and sub categories, ticket requests, email integration, notification, user management, Admin Dashboard, with town adding, client and user controls and management, APIâs for a mobile app and so much more in less than 2 full weeks.
And I really do not think its slop, as a medium full stack developer I have done what would have usually taken several months in a matter of weeks.
I agree it will never be able to compete with an experienced developer in code quality, but no experienced developer would be able to keep up with the productivity rate if you know at least what you are doing.
No more stack overflow questions that you have to wait for hours, and sometimes still canât get the issue fixed.
I can jump from any language and technology stack in a matter of hours and produce descent enough code that would not be utterly garbage. Only the very elite top end programmers could do that⌠until now.
Sorry but no human can keep up with the code it produces when your prompt it correctly. 700+ lines of pretty decent C# code in less than 20 seconds. Who in the world can do that?
