Vibe coded plugins

[Fannon]

Is it necessary to make it work or to get the investers that awesome return of investment they were hoping for ?

The chinese models might spell trouble for it: They are closing in quickly, at least in benchmarks some come close to Opus 4.6. Regarding token cost: Those are often served for like ~3$ per million output token by various hosters (check https://openrouter.ai/models?order=most-popular). They of couse have no incentive or reason to loose money on hosting an open-source model.

So you can see that even today we can do this MUCH (like 10x) cheaper, if we're not going for the very best. Of course China as a state sponsored a lot, so these models could be researched, created and trained - but this is now also out in the open.

[hey212]

They’re burning cash at an unsustainable rate, unless they all magically 10x their revenue. Some of them will go belly up, and whoever survives will be charging orders of magnitude more for tokens.

[_leras]

It’s always been the case that anyone with an idea could create it, if they’re willing and able to invest the time and money to make it happen. The only difference now is AI has truncated the amount of time and money required. Instead of spending $120,000 to hire a programmer for a year, you can now have AI do it in a month for $20. So really it’s no different other than the barriers have been lowered.

Urs seems to be wishcasting for those barriers to be restored. But that’s not going to happen. Genies don’t get put back in their bottles.

I'm not sure any of what you wrote here is true. Not least, not everyone can code all the things needed to build audio tools.

I'm pretty sure a lot of the best modelling is also in house and not shared code. I think it would still take a lot of effort to get to state of the art.

Also AI doesn't got ears...

[IvyBirds]

It’s always been the case that anyone with an idea could create it, if they’re willing and able to invest the time and money to make it happen. The only difference now is AI has truncated the amount of time and money required. Instead of spending $120,000 to hire a programmer for a year, you can now have AI do it in a month for $20. So really it’s no different other than the barriers have been lowered.

Urs seems to be wishcasting for those barriers to be restored. But that’s not going to happen. Genies don’t get put back in their bottles.

I'm not sure any of what you wrote here is true. Not least, not everyone can code all the things needed to build audio tools.

Sure not everyone can code, but everyone can hire someone who can if they have the money to

So you have a guy with an idea for a plugin. He knows nothing about how to code a plugin but he does have the money or can get the money from a bank or from an investor to hire someone who can. So he hires a programmer or a team of programmers and they make the plugin and bring his idea to fruition

That same guy can now save a tremendous amount of time and money in the process by using AI to write some of the code. AI just becomes another tool to do that. Like using Open Sourced Tools or JUCE

I'm pretty sure a lot of the best modelling is also in house and not shared code. I think it would still take a lot of effort to get to state of the art.

You are assuming the goal for all synthesis has to be "modeling". Why is that? Why do Synth Plugins have to model analog synths, acoustic instruments, or anything? Why can't they be their own new thing?. And with all that in house modeling what exactly is it producing in 2026? Yet another modeled Jupiter 8, or ladder filter, or yet another Piano.

Keeping it all in house creates situations that cause developers to have plugin after plugin after plugin that sounds pretty much the same as they never move the ball forward. It's stagnation

As far as taking effort that is one of the advantages of AI it reduces the effort and you can use it as a tool to save yourself weeks and months worth of typing and coding

What AI development offers is a easier path to develop new and interesting things rather than yet another way to send a saw wave through a resonating filter modeled after some 1970s or 1980s synth. Does that mean all of them will? No just like with hand coded plugins many plugins will be dog crap. You just certainly don't need AI to make dog crap Synths

Also AI doesn't got ears...

And? No plugin have ears. The reality is tons of hand coded plugins sound like garbage. The difference is humans behind the project no matter how it was developed do have ears

[IvyBirds]

They’re burning cash at an unsustainable rate, unless they all magically 10x their revenue. Some of them will go belly up, and whoever survives will be charging orders of magnitude more for tokens.

Not really, however the same could be said for search engines in the late 1990s. There were dozens of them maybe hundreds. My brother in law worked for one no one ever heard of and made crazy stupid money as a programmer. After burning through untold millions of VC and never making a dime finally went belly up, yet from that a few successful players did make it and still don't charge for search, and at least one of them is going all in on AI with billions of dollars in the bank to find the effort

So there is no reason to think anything with cost more in the future. Like everything else AI will get better and cheaper

[TechHaus]

Search became a way to display ads. Even Amazon's store search is the third largest ad network in the world.

Ivy, think for a second before you post.

[pekbro]

Yup, that's why I finally cutout all accounts with Amazon, you can not turn that shit off, despite what they say (and even charge you for). Huge scam they are, they even only returned the .10 cent charge to test my CC, after I canceled my prime account, something like 8 yrs later.

True story.

[hey212]

Not really, however the same could be said for search engines in the late 1990s. There were dozens of them maybe hundreds. My brother in law worked for one no one ever heard of and made crazy stupid money as a programmer. After burning through untold millions of VC and never making a dime finally went belly up, yet from that a few successful players did make it and still don't charge for search, and at least one of them is going all in on AI with billions of dollars in the bank to find the effort

So there is no reason to think anything with cost more in the future. Like everything else AI will get better and cheaper

Actually, there are loads of examples of tech services that did not get "better and cheaper", and in fact got significantly more expensive! It's quite a common growth strategy, whereby companies subsidize the cost to the end consumer by selling at a loss, and then after they've captured a large enough market share, jack up prices. The most obvious examples that come to mind include Uber, Lyft, Netflix, Spotify, Air BnB, and Amazon. Following the pattern, Open AI currently spends about $5 dollars for every $1 in revenue.

Regarding comparisons to infrastructure buildouts of the past, the internet/search buildout of the 90s is a good one conceptually, but even when you adjust for inflation does not come close to the amount currently being spent on AI. This is by far the most expensive infrastructure buildout in history, and as such will require a level of revenue growth that has never occurred before, to be sustained.

I have attached Claude's response to the following prompt, "Please compare the amount spent to build out internet and search services during the 90s, adjusted for inflation, to the amount currently being spent to build out AI services." The green bars are the 90s in today's dollars, the blue bars are now.

spending.jpg

[sin night]

I'm cautious about code generated by AI.

I don't use it much at work, but I'm fixing bugs, not developing new features. Whenever I try to ask AI for something about the project, most of the times it doesn't give me the answers I need.


My colleagues writing new features use AI, but they are supposed to control all the code generated by AI. On our big front end project, they say AI often hallucinates so it's not that useful at the moment (while it seems to be doing a better job on micro-services)
At work, I heard stories of a new feature where, at the first iteration of the writing process, AI wanted to delete/drop the whole database. They found the mistake before deploying in the test environment (because they are smart enough to look at the generated code!), but they tell the story as one of those "what the f..." moments. They keep using AI as an assistive technology, they find it useful, but they don't trust it blindly. They still need to write some parts manually.

Also at work, one of our managers (who reached his position through a technical career, so he knows the subject very well) told us his personal experience in his spare time with vibe coding (because he want to have direct experience on the subject) while having lunch. He managed to make a simple application rather quickly, but he applied his programming experience and he somehow guided the AI, as if the AI was a junior programmer, through smaller task, developing and perfecting them before moving on to the next part... but he had the whole coherent vision of the product in his head!
If you're not very precise in what you ask, if the whole process has some fallacies, if you don't know how to properly structure a program... that shows in the output!


And that's still not enough! Sometimes AI changes things for no apparent reason. I wrote this story more than once, about a year ago I gave some math formulas, written as a few lines of pseudo-code (I think 6-7 lines, no more), to a colleague, then he asked AI to turn that into code... then he tested the software but it didn't work and we spent a good amount of time to find out that AI modified my instructions replacing a variable with another one. All AI had to do was to write a few variables declarations, yet AI changed them for no reason (and my colleague didn't check it properly!). Really, I think I could have turned those few instructions into code in the same amount of time required to write the prompt to ask AI to do that...


Being a programmer is not just about coding, there's a whole mindset on how to approach problems, a programmer should be able to see (and possibly anticipate) the issues normal users don't see, to find edge-cases, to find the limits of a process, to design things in a effective way...

The quality of AI-generated code may improve over time (it has rapidly improved in a few years); but I think an experienced programmer has an edge, even if vibe-coding, things are more likely to go smoothly / take less time / being better overall.
I'm not saying a human being is always better than AI (some human beings write outrageous code on their own!), but knowledge and skills still matter.


And there's also an important aspect that's not often considered: when you deliver some code, you should take ownership and responsibility (at least in theory). Can you still comfortably do that, if only AI actually knows how the inner workings of the product?
A programmer doesn't just write code, but certifies the quality of that code (at least in theory).

I can provide a fresh cautionary tale: a few months ago an Italian streamer named Grenbaud created a social network for his fans called "Baudr". He vibe-coded it and spent about 40€. No programmer and no law expert involved. Unfortunately, the service had various serious security issues, quickly ending up with a data breach. It's not funny, especially because he was dealing with personal data, so the GDPR applies...


You need to know what you're doing, especially if you're providing a tool (or the output of that tool, or the AI output) to someone else... It's about the product quality and it's about responsibility.


Then, AI can help in certain circumstances...

[IvyBirds]

Not really, however the same could be said for search engines in the late 1990s. There were dozens of them maybe hundreds. My brother in law worked for one no one ever heard of and made crazy stupid money as a programmer. After burning through untold millions of VC and never making a dime finally went belly up, yet from that a few successful players did make it and still don't charge for search, and at least one of them is going all in on AI with billions of dollars in the bank to find the effort

So there is no reason to think anything with cost more in the future. Like everything else AI will get better and cheaper

Actually, there are loads of examples of tech services that did not get "better and cheaper", and in fact got significantly more expensive! It's quite a common growth strategy, whereby companies subsidize the cost to the end consumer by selling at a loss, and then after they've captured a large enough market share, jack up prices. The most obvious examples that come to mind include Uber, Lyft, Netflix, Spotify, Air BnB, and Amazon.

And none of those are actual technology companies. They just pretend to be. Uber and Lyft would not exist without humans driving cars. They are a taxi service that has an app.

Netflix isn't a technology company they are a media company that distributes movies and TV shows. Those don't exist without human actors, human cinematographers, human directors, human makeup artists, human editors, human sound designers, human composers, etc. All of those people are doing things that were being done 100 years ago

Spotify is a distributor for music they don't exist if it wasn't for humans making music

Airbnb sells vacation rentals in houses owned by other people. Without physical houses they don't exist. They are not a technology company they are a real estate company. They don't exists with out real estate built by humans

Amazon is a retailer with a fleet of trucks and planes, thousands of warehouses, and millions of employees that move physical goods around the world. That is not a technology company it's a retailer that has a website

On the pure technology side Amazon sells AWS services which is SIGNIFICANTLY cheaper to buy into for your business needs than it used to be for the same services

Following the pattern, Open AI currently spends about $5 dollars for every $1 in revenue.

Following what pattern? Open AI doesn't have drivers that will come to my house in their own personal car and give me a ride, they don't distribute movies, TV shows, and music, and they don't sell physical goods that come to my house from a warehouse they own, in a truck that they own. It's rather silly to pretend that Open AI is following the pattern of any of the companies you listed

Regarding comparisons to infrastructure buildouts of the past, the internet/search buildout of the 90s is a good one conceptually, but even when you adjust for inflation does not come close to the amount currently being spent on AI. This is by far the most expensive infrastructure buildout in history, and as such will require a level of revenue growth that has never occurred before, to be sustained.

But that build out is already occuring and will only get cheaper as servers get more powerful and more efficient at less cost. Compare what the servers that powered search engines in the late 1990s to what a server that powers AI today .

I have attached Claude's response to the following prompt, "Please compare the amount spent to build out internet and search services during the 90s, adjusted for inflation, to the amount currently being spent to build out AI services." The green bars are the 90s in today's dollars, the blue bars are now.


spending.jpg

And? What you have is an emerging technology some players will most certainly go belly up just as search engines did in the 1990s. Once they go up belly up the money are they are spending will disappear. That still doesn't mean that any of them will have significant price increases

Compare the costs to build out cell phone networks over the years, now over a trillion dollars in the USA alone. In the early 1990s, adjusted for inflation in 2026 dollars I was paying well over $100 for my base plan and $1 a minute to make a call

I pay less than that now for unlimited calls, I have unlimited texts, and unlimited high speed data, despite the phone companies spending a trillion dollars along the way. I started with company called singular that no longer exists, then I switched to Nextel, again they no longer exist, then I had sprint, they no longer exist. I went with all of those because at the time they offered the best service where I lived and spent major money building out that network

They are spending more to build out AI then they did with search because it will be far more impactful on GDP than search. If the numbers on your chart are correct, and the prompts that you used are logical, what is being spent on AI is a tremendous bargain.

[pekbro]

Actually Amazon's biggest income stems from Data hosting, which 'is' a technology industry.
I was told that by one of the head guys at Amazon a few years back. They lose ton's of
money elsewhere...

*They came to where I work for a convention and I asked the guy how they managed the free
shipping thing.

[hey212]

I was referring to plain old Amazon that sells physical stuff. They spent years operating at a loss, selling books and other items below cost to take market share, and then increased fees for third-party sellers and raised Prime subscription costs. And that is the pattern that Ivy seems to have missed.

[pekbro]

All of that is designed to facilitate the cloud computing thing somehow, interestingly.
Being in Hawaii, shipping quite often exceeds the cost of the item, it didn't make sense
why they would do that, which is why I asked. *shrug

I'm sure way back it wasn't like that, but it is now I guess and has been for years...

[hey212]

Correct, back in the days online companies actually had to buy their own servers to exist. Eventually, AWS grew out of the spare server capacity of Amazon's e-commerce business, but even today their e-commerce business is a much larger share of their revenue. However, it's also true that AWS is a much more profitable business line, which allows them to operate the e-commerce side with lower margins.

[IvyBirds]

Actually Amazon's biggest income stems from Data hosting, which 'is' a technology industry.
I was told that by one of the head guys at Amazon a few years back. They lose ton's of
money elsewhere...

*They came to where I work for a convention and I asked the guy how they managed the free
shipping thing.

Absolutely and they charge less for the same services than they used to