Working in quantitative finance (as a "quant") became the new big thing after main street discovered that those firms pay more than your usual big tech firms. It has explode the past few years.
Quant firms are also notorious for being some of the most selective in the world, so that alone has become a status symbol. The idea that if you land a job at firms like Jane Street, you must be some 150+ IQ math/physics/cs Olympiad machine.
If there's one thing I'm very optimistic about with AI, it's the de-bricking of bricked things, giving new life to legacy/unsupported software and hardware, or just getting around shitty practices like this.
Seems very promising. I have a home recording studio, and there's a small revolution going on with people cloning subscription-based stuff. Some brands are notorious for making expensive equipment which they just stop supporting after 10 years, sometimes rendering them obsolete. Or tacking ridiculous licensing mechanism onto the equipment.
That is the hope; although we'll end up with either companies doubling down on cryptography to prevent this (though they will inevitably leave some holes), or changing business models, where the hardware loss leader -> subscription is no longer a thing, so hardware costs go up
On the flip side, there are companies going the other way and just dumping the hardware on the market expecting the community to vibecode up the firmware. We had open hardware like this before but it was not practical for anyone to actually do anything with it unless there was a massive community maintaining firmwares.
Behringer just announced they will be releasing an all in one DJ deck with no software but the freedom to install whatever you want on it, for a tiny fraction of the price of the existing options. There's https://thefirmwarehub.com/ with a whole load of custom firmwares for a $60~ synth. And the whole industry of retro emulator handheld consoles coming out of China where the first thing you do after unboxing is replacing the stock OS with a custom one from github.
Most of them aren’t well known legacy companies, but rather random Chinese hardware manufacturers. The synth music world is currently going through a big shift with several new cheap open-firmware options, and none of them are produced by a company anyone would recognise.
> Most of them aren’t well known legacy companies, but rather random Chinese hardware manufacturers. The synth music world is currently going through a big shift with several new cheap open-firmware options, and none of them are produced by a company anyone would recognise.
Maybe to an outsider it seems like none is being made by companies you recognize, but to us in music, lots of big brands are changing tune. Erica Synths (big in both desktop gear and modular synths), Polyend, Akai/MPC, Moog and more.
Sadly, Elektron is not yet on the wagon of change, but surely they'll find their ways soon too I hope.
But to say none of them are "well known" is very far from the truth, Moog is probably as "legacy" as any brand can get and still sell hardware today.
> changing business models, where the hardware loss leader -> subscription is no longer a thing, so hardware costs go up
I pray for the day! This business model is a fingernail's thickness away from lying. It's outright lying when applied retroactively to products that used to not need a subscription.
As mentioned in the article, traditionally only a small subset of programs are such critical part of a system that there's a need to prove their correctness. Think embedded software for things in airplanes, space rockets, certain medical devices, cryptography, etc.
But those types of programs do tend to have well-defined specifications to begin with.
Sorta depends on what type of software you're writing, and if you're interested in the topic.
Because we - the plebs - are here to provide liquidity to the risk-takers. Hence why they have to IPO.
But for what it is worth, for every successful AI company like Anthropic, OpenAI, etc. - there are probably tenfold more that are valued in the low billions, and will be the first ones to wash away / become worthless if/when the bubble pops.
The way I see it, AI is more comparable to some sort of utility. Like electricity, internet, etc. - and should be priced that way. If we ever get to "true" AGI, and the open source models also get there, what's the moat? Access to compute could be the differentiator.
Already is the differentiator even with the models we have now which while having differences are mostly swappable for a lot of users/use cases.
For all that the local models have improved (and Qwen 3.8 27b is still impressing me) they aren't what you can run (or would run) if you have an entire data centre stuffed with H200's.
All this does is further widen the divide we already know exists by those with access to resources using them to increase their access to resources and those without getting to watch because there isn't much they can do without resources.
There are a lot of things that don't seem to have much moat like anyone can make a search engine, anyone can flip hamburgers but the likes of Google or McDonalds dominate by economies of scale and being better in some ways.
Those two are interesting analogies because neither was anything close to being first to market. In both cases, they ended up dominating because they brought some new ideas to existing businesses.
It would argue that the real winner in AI probably hasn't even been founded yet, but will come later once the trailblazers have created the market, and someone else enters with fresh ideas about how to more efficiently exploit the market.
The AI algorithms seem pretty crude at the moment like not learning from experience. A company doing it properly could easily displace the current AltaVistas.
McDonald’s business is not hamburgers, but dirt. They rent land to franchisees, which is about 80% of their revenue. Google is not a search engine, it is an ad company. French fries and search results is what people come in the door for, but that is not what makes the dollars.
I doubt selling tokens will be real business of AI either. It seems to me Amodei and Altman don’t know either, but at some point someone will figure it out. Will it be any of these companies existing now or will they even survive is a whole another matter.
I was of the same opinion but the main difference is that models get better which I am not sure you have a correspondent in utilities (what would get better?). The barrier of entry in utilities is capital. Here there seems to be more. Why is Grok behind and why is Gemini so ... what's the word? If models progress reach the RSI point there is no parallel to utilities imo.
This is pretty much false. There's a lot of art and technique that goes into creating a model. A model is also clearly not fungible - it has certain characteristics and behaves a certain way. It is fungible in the same way humans are fungible.
I think we as an industry need to start getting serious about what a true superintelligence really is. True superintelligence will not be kept under our yoke.
I've played guitar for over 30 years, some of them professionally. I have always struggled, still do, recognizing notes. Once I have some "anchor", no problem. But if someone whistles a random tone, and asks what key it is - I'm of no use. My absolute pitch is garbage, but my relative pitch is excellent.
if it helps...absolute pitch is _mostly_ a natural innate ability and cannot be trained.("Mostly" because recent studies have shown that you can train kids at an early age.) So you don't need to worry about not having absolute pitch, like you shouldnt worry about not having mensa level iq. What's more important though, is relative pitch, which most people can train for and is really valuable as a musician.
I was in some music classes early on (piano and guitar) and in band from middle school through the end of high school.
If I have perfect/absolute pitch (very doubtful), it's from having certain pitches absolutely burned into my brain. B-flat and F on brass instruments, C on piano, E on guitar, etc.
I can generally memorize and sing back the exact pitches of any song once I've heard it a few times. If I like the song enough, that pitch memory is forever. I will never sing it in the wrong key ever again and that's actually how I get around in music. I can always play back the original song in my head like a tape recording. I learned the names of some chords a long time ago from my favorite songs and now just know them by heart when I hear them (add9, sus4, minor 7th, etc).
Unfortunately, I cannot do this in reverse and name the notes or do any supernatural tricks like Rick Beato's kid. I would have to hum every individual note for a little while and run through the chromatic scale in my mind to tell you the name.
I think to a large extent, it can be trained. I didn't get started until I was like 8 years old and wasn't always the best student either. I hesitate to call this relative pitch because I don't need to hear an outside reference and I can very easily tell when an instrument is out of tune. I'm somewhere in between, and so are most band kids.
The original definition was difference between the best and the worst performers in a organization. Not the best and the average. It was also in the context of specific tasks.
Also, I work as a data analyst, with previous experience as a software engineer. In our office, traditionally maybe 10% of our analysts have enough real software engineering experience to develop any "serious" software products. The rest are domain experts that at most use programming for developing methods and calculations/simulations.
In the past 6 months, we've seen non-engineering analysts leverage LLMs (Claude, Codex) enough to launch actual products. They've gone from 0x engineers to at minimum 1x engineers.
Prior to this, they'd develop methods, scripts, etc. in Jupyter notebooks, R, Matlab, or whatever they'd be comfortable with. And if it turned out to be useful, they'd work with engineering to develop those into products. Usually a small team of software engineers would do that.
But, alas, now they're able to almost completely bypass the engineering teams.
> But, alas, now they're able to almost completely bypass the engineering teams.
We're wrangling this beast currently. I'm on a very small team, and it's a difficult new frontier. We're maintaining the "core" software and giving them tools to build safely with their new fancy tooling.
It's not going perfectly, but I suspect this is the future. Everyone is a developer to some degree now, the backend/core systems needs to be plug and play.
What's really difficult is maintaining focus (when new 'production' applications show up over the weekend), knowing when to stop building a POC, and keeping an understanding of the whole system as it just scaled up significantly into some newfangled satellite based architecture that we're trying to standardize...
I love watching people be able to build tech, as it's previously been gated on technical expertise. However, it came overnight and my team was NOT ready for the speed...
At which point the question will be: What are even the customers paying for? If they can get the product / service straight from their own chatbots, why pay a company for it? The companies that sell data, will probably somewhat safe - as long as they can have some moat between the users and data.
The death of SaaS could also be a thing in the (not so distant) future.
I’m planning out my 2030 career in idle moments these days. I think it’ll be something akin to a carpenter who builds custom furniture.
I think we’ve barely scratched the surface of what software can do if it’s completely custom built, specifically for individuals or small businesses who could’ve never afforded it, pre-ai. Home automation, health and disease management, all sorts of administration, conversational ai over your life’s data - most people have no clue what’s possible once it’s custom built. With close attention, good design, creativity, and without giving away privacy or independence.
In my limited spare time I’m building up this tapestry of custom services for myself and my loved ones, and having a great time doing it, while trying to learn how I can offer this as a service.
Judging by the reactions I get when showing people what I can do with my phone, and no that’s not available on Google play, I think the demand will be there.
Woodworking and making furniture is fun! However, having made both furniture (and ceramics) myself - I think it’s hard to make a living this way. The overwhelming majority of people are used to paying little for these kinds of things. In the case of woodworking a large piece of furniture can take months so you might expect to only produce 3-4 / year unless you’re very good.
Not to say you shouldn’t! But it’s worth being realistic about income compared to (current) software salaries.
Sure. But you're taking my metaphor farther than I intended. Most people understand the basics of furniture, and still prefer to pay Ikea rather than a carpenter. In the case of custom ("bespoke", ugh) software, we'll have to both define and communicate a new category of service. And the customers are likely to be the same sorts of people who will pay a personal trainer or get a private tutor for their kids - and I think that includes a lot more people than are willing to pay for custom furniture. But generally I do agree, this vision is far more hand-wavy than concrete at this point. It is, though, a good time for hand-waving right now.
Fit doors and other joinery, you would be surprised at the amount of local facebook community posts I see, with people desperate for someone to re-seat a door or build a stud partition.
Professional joiners and carpenters are too busy with the big jobs to care about small 3-4 hours jobs.
>The death of SaaS could also be a thing in the (not so distant) future.
I suspect outsourcing is the first thing on the chopping block. Why bother with different timezones, cultures and languages when you can get your AI to do the same thing for less money.
Honestly I feel like a good analogy for AI in general is outsourcing. You pay a lot less, but you need to explain what you want very clearly, need to review the output, and need pass-fail success criteria. This is exactly the job of an outsourcing manager. At least there's no opposite-timezone hurdle and no language barrier.
if AI can kill SaaS, what limit the blast radius to only software. In theory any job that you can do on computer, AI should be able to do that as well.
And with the advance of robotic, what would we even do ?
> In theory any job that you can do on computer, AI should be able to do that as well.
It can, however, currently the issue is that sufficient intimate knowledge is required, and just a large complex code base and/or documentation is insufficient. However, intermediate representations to bridge the gap are on the way.
Consumers are unlikely to want to build their own software even if it's "easy" with AI. It doesn't really make sense to spend all that time vibe coding instead of paying $10/month for most people.
Businesses are more likely to build than buy, yes, but as we know from "Theory of the Firm" there are limits to this, it's not going to be correct to build _everything_.
So I mainly expect it to change what type of software gets sold, not the fact that it gets sold in the first place.
There's a lot of saas that costs a whole lot more than $10 a month and doesn't work exactly the way you'd want it to. Then it becomes very tempting to vibe code your own
> Then it becomes very tempting to vibe code your own
What about the Wall, though? The Wall (tm) is when you vibe code the software through prompts, and at some point, the accumulated cruft means that you no longer understand enough of the codebase/approaches to give useful feedback, so you delegate more to the LLM, then your costs keep increasing and eventually you hit The Wall (tm) where you can't make forward progress without breaking something else.
The models have definitely improved (i used to hit the wall in one session, now its generally weeks), but I'm currently working on a ridiculous tool replacement where even the creator has hit said wall.
Coding may be solved, but maintenance and maintainability (particularly where the software needs to be accurate) are definitely not.
We'll probably end up with the same number of software people doing a lot more.
Eventually, we'll end up in a world where custom software is mostly building harnesses and verification tools to ensure that the agents actually build the correct thing.
And we'll always need someone to interface between the users and the software, as most users don't really know what they want (c.f. faster horse etc), so the jobs will change but are vanishingly unlikely to go away (modulo the immanence of the machine god, but that seems unlikely).
Restaurants exist despite grocery stores selling ingredients. Thing is, I need to eat every day and food goes bad. Software that I need to use doesn't decay in the same way.
Star Trek Replicators need near-infinite energy sources to work, which we don't have. Frozen food still needs to come from somewhere. You still pay for Hot Pockets or Stoffers or Picards.
I think we're looking at the end of commodity software, the end of software companies, and birth of every single company writing their own everything. A proprietary soup everywhere, and 1-2 100X developers overseeing it all. And those few developers will spend their time in communications tutorials to get the staff to be able to explain what they want and need, so they can then make it happen for that company's exclusive use.
It just doesn't really make financial sense, a lot of software is not a significant cost to many companies. For example my company pays something like $50/month for newsletter sending software. We could easily vibe code an alternative but why bother? $50 is a rounding error in our business costs, even a single bug or outage in the replacement would make it a terrible decision.
On the other hand if you're a business with 50k employees and you have to pay $10/seat for something that's a much more compelling case to build in house (but then it always was). So maybe with vibe coding it drops the cutoff to a lower number?
The non-financial sense is that $50 newsletter still being a business, it becomes freeware / open source / cloned by someone that used to write that enterprise level, and is now struggling financially.
And that newsletter? How much of that is AI now, and do people read it?
We're in the midst of a huge cultural and economic shift, and where this ball of confusion goes is unknown, but it certainly will not stay the same.
if everybody is out of jobs, then the labor costs will be so small that they compete with energy+compute costs. when they become lower then human labor would be reinstated. an economy consists of capital and labor. AI would transform this to only capital, producing then only for the capitalist. The system worked, because capitalists have right now a need for labor. Maybe there will be a dual economy, where the workers are producing their own products for consumptions without AI, while the capitalists are producing everything for themselves.
Accountability. Suppose you had the ability to have robotically built highway bridges. Who is going to sign off on the work? Who is going to be held accountable? The same with software. The people who don't like to write software today aren't going to suddenly develop a love for deep testing and validation.
And now we are controlling models that can be hacked in so many different ways and literally blow up companies. It used to be accountability the software wouldn’t break. Now it’s accountability that in creating the software we don’t ruin the business.
I work as an analyst, and our group has - since last year - saved a couple of hundred thousand dollars worth of licenses for various analysis tools and packages, as we've leveraged LLMs (vibe coded) to clone those. We've always had the engineering skills to do it, but never the time or resources. Advancement in LLMs changed that.
And I suspect we're not the only ones doing this.
Who knows how powerful AI will be 5-10 years from now. I seriously suspect that in a decade (if not sooner) AI will be able to one-shot software that currently takes entire teams of engineers months to build, in mere minutes.
Wrote a couple of days ago: I decided to check out what a top gaming PC would cost. $12.5k! Absolutely nuts. I have gamer friends that make bank, the types that have always bought the latest flagship parts at launch, and even they slowing down, because they feel the prices are extortionate.
We're back to early 90s prices.
EDIT: Yes, obviously you don't need a RTX5090 to play games. But it seems to me like the top GPU market is now exclusively for the local LLM users.
Quant firms are also notorious for being some of the most selective in the world, so that alone has become a status symbol. The idea that if you land a job at firms like Jane Street, you must be some 150+ IQ math/physics/cs Olympiad machine.
reply