I think this has always been the end result - when the user can enter a business problem and it spins up a site, and they don't have to think about the hosting, database, app, etc.
Its not for those uses cases. though. THink about squarespace/wix hosting all the company landing pages, restaurant menus, daycares. This will replace it.
> And it will be good enough for 90% of use cases.
I think it’ll just raise the automation bar that much higher. In other words, it will be good enough for 90% of yesterday’s use-cases. But tomorrow’s? I’m bearish.
I imagine the amount of onerous bureaucracy when interacting with any entity is going to explode if they expect you to do it in an automated fashion, for both private companies and the government. Especially as roles are cut because they can be “automated”. Except only 80% of the workload can be reliably automated, the other 20% requires a long tail of effort.
I mean, with a Cloudflare API key you can pretty much already do that. But keeping your data straight and putting in the right authentication guards, etc. are all still where the business problems meet technical and the AI is not enough unless the agent is both architect and customer.
That's what I do but normal people don't know what any of those terms mean so their daily AI app does that shopping for them and they don't need to get an "Appy Key".
I don't think this provides a lot of value for more complex applications. It just lowers the bar for getting started.
Once the app gets a bit more complex, it won't be that much different to any more traditionally vibe coded application (code, git/pr, test, deploy, ops). It only removes the barrier to get started, but doesn't solve any of the real problems. Problems like AI slop, code rot, bad architecture, security issues, data loss/breach because of bad code, and so on.
That's my experience as well. Some folks like to dabble, but for the most part they have other stuff they want to focus on, they don't want to spend the time to ensure it's good and it works.
I'm not saying an agent couldn't make sure it's good and it works, but there are a lot of aspects to that: It's compliant, it's secure, the right data is stored, the right compromises are made for sometimes mutually exclusive business priorities, UX is decent, it's responsive, it scales, hosting is economical, it has the right tests, monitoring for the right risks, and so on and so forth. A lot of devil in the details, and the context required to know what to focus on is rather vast.
If anything, I see this work getting harder, since the discipline around minimal scope is getting weaker, leading to a complexity explosion affecting all of these aspects.
It usually comes down to how successful a business is. Just starting out with little to no funding? They'll want to dabble themselves. But if there's money, it can and will be exchanged for time. Easier prototypes and MVPs means faster business model validation, which means higher willingness to invest. I do see this happening.
Will we always need to know programming languages for doing this? Maybe not, but it doesn't matter. If you make a machine do stuff for a living, you're a programmer. How exactly you do it doesn't affect that in my mind.
The Saaspocalypse was a wet fart that never went anywhere. People aren't using these AI tools to kill or replace SaaS at their jobs as much as HN wants us to believe.
The argument there wasn't that everyone everywhere would use individually customized software* instead of packaged SaaS; it was that, because AI greatly reduces the cost of building an internal app or service, it improves enterprises' BATNA when dealing with SaaS vendors, which means surplus will shift from producers to consumers. It's not obvious to me that that's not happening, since it may have been masked by a general increase in automation since AI also synergizes significantly with SaaS. (The Stripe Economics post gets into this a little, though it's mostly less curious than I'd like about why the "SaaSpocalypse" didn't materialize. Of course, Stripe has an economic interest here.)
* Some people do think this (e.g., <a href="https://sockpuppet.org/blog/2026/09/25/what-even-is-an-os-now/" rel="nofollow">https://sockpuppet.org/blog/2026/09/25/what-even-is-an-os-no...), but I remain somewhat skeptical that this'll be true for the most economically significant applications. This is not because I'm an AI capabilities skeptic (I'm not, I'm more of a doomer), but because even in a world where AI is designing and implementing all the software, there are probably still significant benefits to standardization. I see it as sort of analogous to compilers; we might get a world where compilers are both used and maintained almost exclusively by LLMs, but we probably won't get one where compilers stop being used or maintained at all because LLMs write everything in assembly.
In stuff like wikis and ticket systems what I see is people now just making their own little tools around the ones we already use, this arguably just sements their position even deeper.
Also I dont get the inhouse argument, you dont really save anything. So instead of paying X to SaaS, you now propose to pay X to anthropic instead AND maintain said SaaS... what? how is that better, that just shifted the cost to another company and you got the burden of maintenance and responsibility.
Honest question for devs out there... is anyone worried? I don't see enough frank conversation happening around this subject. Timelines. How long until salaries fall? Is it too early to stalk thinking about back up plans? Am I crazy?
Coding is probably solved. Honestly it’s probably going to be 5 years of runtime starting now depending on industry lag and uptake.
A team I was managing spent a whole year converting government forms into a web app. It took a while and there was a lot of discussion with the end users and a bunch of back and forth (keep in mind this is government software with a lot of onerous legal requirements). Then about 4 months ago, we just fed Codex the forms, along with the UI design and the legal requirements and it spat it out in 1 hour. Was it perfect? No. But our iterative loop became make the web app, show it to end users and then incorporate feedback and completely cut out the UX team, product etc. We were able to complete the conversion in 2 weeks with one guy (something like 150+ government forms/applications). For most simple to medium corporate development work it’s basically over.
We're in a similar boat. Entire eng team went full AI about a year ago, and honestly we're struggling to find things to do. Features that would take weeks, take minutes.
I just don't see how we keep going like this without layoffs. The company cannot responsibly continue paying for this many engineers, when half, or less, could do the same job.
> honestly we're struggling to find things to do.
I have been the lone developer on a project for quite some time and was already feeling that way. Yet, recently found out that a second is being added to the team and a third is to come. I guess I'll enjoy what is effectively a paid vacation before it all comes crumbling down.
Serious question from a consumer of software: if software can be improved so effortlessly now, why isn't it getting better?
Like, why is it all still so buggy? Why does no website seem to consider 96 PPI displays in its design? Why don't we have ostensibly useful features like undo trees everywhere? Why is everything still a RAM hog? Why does the proportion of websites that pay no heed to accessibility seem to be going up rather than down?
I'm told we're living in the future, but it sure doesn't feel like it.
Lack of information. When you are on the building side you become blind to its faults. Consumers almost never tell you what isn't good about it.
> why is it all still so buggy?
Also lack of information. I haven't seen any bugs in years in the things I've worked on. LLMs have closed every last obvious testing gap, squashing any outstanding bugs there. That's not to say it is all bug-free now, but what bugs users may be experiencing aren't being reported.
AI is good at building software, but it is still not good at consuming software like actual consumers do.
For the same reason that increasing headcount never really solved the bug problem.
Flawless software isn't the goal, ever. Feature development is almost always the primary goal, and fixing bugs and performance issues is only done if the bug is bad enough.
I don't think you're wrong overall, and it jives with tripleee's reply, but I don't think wanting end-user software to be less buggy than it currently is is asking for "flawless" software.
Because the goal isn't to make the future better for humanity. It's to make the future more addictive for humanity. Big difference. More clicks, more ad views, more engagement, not necessarily more pleasant. I mean, that's obvious.
I live in the future when it comes to my hobby of music making. I make my own little software tools that I did not have the ability to make 2 years ago. Little tools that are too specialized and specific for my obscure taste to have much wider economic value as tools in themselves.
Claude is also the greatest obscure book recommendation engine I have ever found so I have a lot more interesting books to read than in the past.
Outside of this, I basically live in 2019 but everything is more expensive.
I worry this is the future. It doesn't really feel much different other than things getting more and more expensive.
If your company isn't throwing you into R&D right now everyone is looking at layoffs. You are no longer in an environment where doing "the same job" will keep you afloat.
I've said it before but I'll repeat it here until someone changes my mind: jobs in IT are safe until using AI gets its equivalent of "no one ever got fired for buying IBM". Until there can be financial and legal responsibility for when things go bad, people are going to be needed.
Once it's so good AI labs will be happy to accept liability, it's game over. It was never about code quality.
You're not crazy. I think there's still room for the top 5% of devs working on genuinely hard technical problems.. but I don't see a bright future for the rest.
No. Saas hedge is to offer solid api + mcp for agents to use whilst still maintaining their own products alongside. That's it, that's the whole thing.
Are you going to trust a vendor who has been running for years with your business critical operations or brad_69's awesome saas skills and agents? Do you think these businesses are going to shift their whole product to be in Altman's walled garden when they don't need to?
There’s going to be a lot of chaos at some point but I think that “hybrid builders” (scientists, sales people, product developers, etc). true specialists, and new opportunities in hw/robotics/training/ will soak up a lot of talent, and be rewarding to work in. Feels very sadly k-shaped.
Software engineering on one end is going to be more like finance or medicine where the cost of poor performance is actually higher than the risk that they wont be worth what you pay them (ie there isn’t any amount of pay where it’s worth it to hire them). Then on the other it’ll be like knowing how to use PowerPoint. All the new tech categories have too many bottlenecks and challenges beyond just producing code to consolidate like that.
Slapping up a slop website is no threat to professional tech folk.
When they have full DevOps, with multiple environments for testing, compliance with security best practices, can do database updates and data migrations, move the app smoothly between different platforms, streamline costs, and scale to at least hundreds of thousands (if not millions) of users without crashing and burning... then go ahead and worry. We aren't anywhere near that.
On the other hand, if you are purely a coder, and cannot run a project that does all of the above... maybe start your worrying earlier.
We aren't, but the AI Labs no doubt are. They just can't release it right now because their bread and butter is still developers burning tokens, so there is still a business problem to solve.
Salaries won't fall, they'll go up. The number of roles is what'll fall. SWE is going to become a giant game of musical chairs, and people who can't find a seat will get pushed into human facing roles ("product" / "technical sales" / "customer support" ).
It'd be hard to compete when there are too many sellers like you. In many businesses, data is the moat and network effects keep making them more valuable over time.
I think technically building this stuff is now easier for the masses. But knowing what to build, how to describe it, how to test it, figuring out how to understand the users -- that's the hard part, always was and I'll be employed doing that for a couple more years at least. But after that, I see this all going towards building for agents, building for users will be a thing of the past. Then I think its not as easy for people to relate to that, probably want a machine designing for other machines. UX and taste don't really matter much at that point.
No, you should be thinking about backup plans, because you should assume you’re not going to walk into another career (I am not suggesting that you are).
- People in other industries actually have specialist knowledge (this is something IT people forget)
- AI is coming for those jobs too
- and so are lots of other people, not just displaced techies but from any number of other industries.
Thinking about it in advance should include training yourself up for other sectors.
If AI is successful, it will be a very different world of jobhunting. If it’s unsuccessful there will still be job losses for a while because of overspending, and because of the wider economic issues.
It is not clear to me that it is yet at a level of success to cause massive tech job losses (it's just causing large-scale misery).
It's also not clear to me that the level of success in the tech industry is truly being repeated anywhere else. There are industries where generative AI is actually a liability (the law, finance) and where it is increasingly viewed as repugnant (design, the arts, to some extent photography where deep learning noise reduction tools are OK but the culture is swinging hard against other generative uses).
I am also not convinced it is "objectively" successful in the tech industry, considering the extent to which most token pricing is subsidised and people's reluctance to demonstrate ROI even on subsidised/bundled tokens.
I'm not claiming it is unsuccessful. Just that I don't think the job apocalypse is here yet. Though this product (ChatGPT Sites) is very indicative of a coming tech job apocalypse in pure software services.
It's certainly at a level of success where people should start thinking about a "Go bag" on an employment level. Keep your CV up to date, work on gaps in non-tech skills, consider other qualifications.
On one hand, I am incredibly tired of seeing the absolute train wreck of waste that is suave founders taking millions and blowing it on hundreds of personnel who do practically nothing— especially having seen what a single person can do with these tools.
On the other hand, death of a livelihood in a social structure that doesn’t care about you.
We’ve always had pump-and-dump trash. The terrible stuff loses and the good stuff wins. I don’t see why that would change.
The companies that ditch the management structures of yore and successfully build better software faster with the new tools will come out on top. I don’t think these companies will replace ALL saas subscriptions with in-house solutions, because one huge benefit of paying someone else to hold that bag is liability and regulatory overhead.
They’ve both been working on this for a while, I’m surprised they still haven’t launched.
Domains, auth, and databases are probably the stickiest software products of all time. I suspect it’s a combo of security concerns and the risk of eroding goodwill among developers (their most important customer base rn) if they move up the stack too quickly. Once you know that the goal is to become cloud vendors, train on/compete with their own customers, and own the entire software stack e2e it’s hard to really feel grateful that they’re dragging it out but planning on doing it anyway.
For those of us working in infra/SaaS outside these companies it’s pretty clear that the only viable path that doesn’t involve getting cannibalized is training your own agents/models. The new coding agent-infra-data business model is a path towards full commoditization and undifferentiated prompting in 3-4 vertically integrated walled gardens. If you ever start making real money on pure software infra they’ll just be able to eat you alive by implementing something similar, training on their own tools rather than yours, and integrating it into their stack (which all of your customers are on already).
Also, you should only ever use $20-200/mo subscriptions on work you want them to train against or you don’t mind automate yourself out of. Think a little bit about what you’re teaching them to do when you use their products, especially if they’re the primary interface you’re working in. People are going to start caring about this a lot when the AI companies feel safe enough to begin the “extinguish” phase of AI coding. Train your own models!
It’s really only an existential problem for pure-software businesses/developers.
There’s a good chance that the next step after open weight models is something that makes it easier to post-train existing models on new workloads without running an entire pipeline just for yourself. Somebody has to actually build it though.
> I do not see enough people discussing this and trying to position their businesses in a way to avoid destruction at the hands of the labs.
What can most even really do about it? Assuming they pull off what they're saying they will: it won't matter what you do, what vertical, what moat you think you have. It will be a concentration of capital and power we've never seen before, and frankly that's terrible for the world.
We need to build collaborative post-training pipelines to augment models with new custom capabilities without catastrophic forgetting/excessive loss of capability in other domains.
Currently what makes this too difficult for anybody but frontier labs is the lack of access to the full distribution of workloads/data they use to RL multiple separate envs/evals without regressing more than they advance in general capabilities. Because they are the primary buyers/builders of that stuff, they have no incentive or reason to allow anybody to replay or resample it but themselves. And it's all so very expensive to do in aggregate so there's not really demand for other product shapes except more openly available, collaborative/bundled RL that you could plug your specific workloads into (who the frontier labs would obviously not want to support with their business).
However, if someone were to build a collaborative rollout platform that you could use to train private workloads (ie create useful IP that doesn't just become profit for other labs/come from what they already can do), the upfront investment would be amortized over the very large number of potential buyers once it gets into the 5-8 figure range, who essentially have no other choice if they want to remain competitive in the technology industry.
Until openai/anthropic ipo the concentration of capital/spending and ndas/loss of employability is too concentrated for the best researchers to really do this without rocking the boat. And a lot of also-ran ai/saas have the same risk due to the lack of capitalized acquisition opportunities or AI vendors to partner with.
If the plan is ultimately to drive you out of business if you ever build anything profitable with their products, and monetize your knowledge without fairly rewarding or explaining their intent to do so, you might as well defect early and build what you inevitably would need anyway.
I think that overlooks requirements like "absolutely ensure the end-user has to have seen and agreed to this exact text before billing them, because otherwise it's a massive fine."
Pessimistically, companies will ignore that and pick the cheapest solution and accept an unknown failure-rate that they assume will be tiny.
Optimistically, they will demand "the AI you just demo'ed, except fix whatever bug so that it never screws up." In that case, SaaS will not be about "training your own LLM model", but about investing in harnesses and rule-engines and GOFAI that constrain the actions of a not-necessarily-custom LLM which is acting as the consumer's UI or concierge. In other words.
It’s mainly mucus, from a piece of mouse colon that was cut open and gently flushed out, leaving the inner mucus layer attached. The actual mucus is black here by default but there is another channel you can turn on in the sidebar to see it.
On top of it we added bacteria (green) and bacteria sized beads (red). On the bottom are colon lining cells (blue).
The inner mucus layer of the colon was generally thought to be impenetrable to bacteria, serving as a way to protect you from the trillions of bacteria in the gut.
It turns out that actually, it’s really easy for bacteria to swim through mucus by using flagella. We found some bacterial mutants that swim through mucus (the green ones). The red beads serve as a control marking the top of the mucus since they can’t get through the mucus.
Find this and other images in my paper: <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC11784811/" rel="nofollow">https://pmc.ncbi.nlm.nih.gov/articles/PMC11784811/
The real question is then - if they can get through the mucus at any time, why don’t they? And there are various theories about this - which I’m happy to elaborate on this you want. The basic idea is there is a risk whenever you cross the mucus (you get closer to the immune system) and mucus actually feeds and tames bacteria over time by providing them with various sugars and things like that. But if they get desperate/starve, then they will cross.
Mucus is still a mystery - there are substructures within mucus, take a look at this image for example:
bananaflag · · focus · HN ↗
copperx · · focus · HN ↗
Fergusonb · · focus · HN ↗
Vibe coding full circle
And it will be good enough for 90% of use cases.
avs733 · · focus · HN ↗
It is likely better in some ways and worse in others...and few of hte people involved will have the understanding to tell the difference.
carlosjobim · · focus · HN ↗
bushbaba · · focus · HN ↗
camdenreslink · · focus · HN ↗
datsci_est_2015 · · focus · HN ↗
I think it’ll just raise the automation bar that much higher. In other words, it will be good enough for 90% of yesterday’s use-cases. But tomorrow’s? I’m bearish.
I imagine the amount of onerous bureaucracy when interacting with any entity is going to explode if they expect you to do it in an automated fashion, for both private companies and the government. Especially as roles are cut because they can be “automated”. Except only 80% of the workload can be reliably automated, the other 20% requires a long tail of effort.
axus · · focus · HN ↗
nathanfig · · focus · HN ↗
pampas · · focus · HN ↗
andix · · focus · HN ↗
Once the app gets a bit more complex, it won't be that much different to any more traditionally vibe coded application (code, git/pr, test, deploy, ops). It only removes the barrier to get started, but doesn't solve any of the real problems. Problems like AI slop, code rot, bad architecture, security issues, data loss/breach because of bad code, and so on.
pampas · · focus · HN ↗
OtomotO · · focus · HN ↗
An average customer of mine hasn't got the slightest interest in dealing with this stuff.
Sure, they show me llm generated prototypes these days of how they envision a certain feature, ui and ux wise.
But that's great, because even with AI I save time iterating with them.
But for the actual programming and maintenance they still come to me.
Why? Because THEY ARE NOT INTERESTED IN IT!
fhd2 · · focus · HN ↗
I'm not saying an agent couldn't make sure it's good and it works, but there are a lot of aspects to that: It's compliant, it's secure, the right data is stored, the right compromises are made for sometimes mutually exclusive business priorities, UX is decent, it's responsive, it scales, hosting is economical, it has the right tests, monitoring for the right risks, and so on and so forth. A lot of devil in the details, and the context required to know what to focus on is rather vast.
If anything, I see this work getting harder, since the discipline around minimal scope is getting weaker, leading to a complexity explosion affecting all of these aspects.
It usually comes down to how successful a business is. Just starting out with little to no funding? They'll want to dabble themselves. But if there's money, it can and will be exchanged for time. Easier prototypes and MVPs means faster business model validation, which means higher willingness to invest. I do see this happening.
Will we always need to know programming languages for doing this? Maybe not, but it doesn't matter. If you make a machine do stuff for a living, you're a programmer. How exactly you do it doesn't affect that in my mind.
OtomotO · · focus · HN ↗
That and you need to be able to even ask the right questions.
These are people that have NO IDEA about how computers or their software works.
Sure they can let it be explained by an LLM but unless they thoroughly study these things, they simply can't know which questions to ask.
nozzlegear · · focus · HN ↗
<a href="https://www.stripeeconomics.com/p/the-saaspocalypse-was-more-like-a" rel="nofollow">https://www.stripeeconomics.com/p/the-saaspocalypse-was-more...
ameliaquining · · focus · HN ↗
* Some people do think this (e.g., <a href="https://sockpuppet.org/blog/2026/09/25/what-even-is-an-os-now/" rel="nofollow">https://sockpuppet.org/blog/2026/09/25/what-even-is-an-os-no...), but I remain somewhat skeptical that this'll be true for the most economically significant applications. This is not because I'm an AI capabilities skeptic (I'm not, I'm more of a doomer), but because even in a world where AI is designing and implementing all the software, there are probably still significant benefits to standardization. I see it as sort of analogous to compilers; we might get a world where compilers are both used and maintained almost exclusively by LLMs, but we probably won't get one where compilers stop being used or maintained at all because LLMs write everything in assembly.
tripledry · · focus · HN ↗
Also I dont get the inhouse argument, you dont really save anything. So instead of paying X to SaaS, you now propose to pay X to anthropic instead AND maintain said SaaS... what? how is that better, that just shifted the cost to another company and you got the burden of maintenance and responsibility.
ameliaquining · · focus · HN ↗
tripledry · · focus · HN ↗
The SaaS to inhouse argument is just something I've heard that came to mind when reading this thread.
jr3592 · · focus · HN ↗
dyauspitr · · focus · HN ↗
A team I was managing spent a whole year converting government forms into a web app. It took a while and there was a lot of discussion with the end users and a bunch of back and forth (keep in mind this is government software with a lot of onerous legal requirements). Then about 4 months ago, we just fed Codex the forms, along with the UI design and the legal requirements and it spat it out in 1 hour. Was it perfect? No. But our iterative loop became make the web app, show it to end users and then incorporate feedback and completely cut out the UX team, product etc. We were able to complete the conversion in 2 weeks with one guy (something like 150+ government forms/applications). For most simple to medium corporate development work it’s basically over.
bushbaba · · focus · HN ↗
dyauspitr · · focus · HN ↗
bushbaba · · focus · HN ↗
jr3592 · · focus · HN ↗
I just don't see how we keep going like this without layoffs. The company cannot responsibly continue paying for this many engineers, when half, or less, could do the same job.
win311fwg · · focus · HN ↗
I have been the lone developer on a project for quite some time and was already feeling that way. Yet, recently found out that a second is being added to the team and a third is to come. I guess I'll enjoy what is effectively a paid vacation before it all comes crumbling down.
creata · · focus · HN ↗
Like, why is it all still so buggy? Why does no website seem to consider 96 PPI displays in its design? Why don't we have ostensibly useful features like undo trees everywhere? Why is everything still a RAM hog? Why does the proportion of websites that pay no heed to accessibility seem to be going up rather than down?
I'm told we're living in the future, but it sure doesn't feel like it.
tripleee · · focus · HN ↗
win311fwg · · focus · HN ↗
Lack of information. When you are on the building side you become blind to its faults. Consumers almost never tell you what isn't good about it.
> why is it all still so buggy?
Also lack of information. I haven't seen any bugs in years in the things I've worked on. LLMs have closed every last obvious testing gap, squashing any outstanding bugs there. That's not to say it is all bug-free now, but what bugs users may be experiencing aren't being reported.
AI is good at building software, but it is still not good at consuming software like actual consumers do.
copperx · · focus · HN ↗
jr3592 · · focus · HN ↗
Flawless software isn't the goal, ever. Feature development is almost always the primary goal, and fixing bugs and performance issues is only done if the bug is bad enough.
creata · · focus · HN ↗
FeteCommuniste · · focus · HN ↗
Like the old quip that a cup of coffee lets you do stupid things faster.
vouaobrasil · · focus · HN ↗
aryehof · · focus · HN ↗
rl3 · · focus · HN ↗
<a href="https://www.youtube.com/watch?v=vzm6pvHPSGo" rel="nofollow">https://www.youtube.com/watch?v=vzm6pvHPSGo
lutanista · · focus · HN ↗
I live in the future when it comes to my hobby of music making. I make my own little software tools that I did not have the ability to make 2 years ago. Little tools that are too specialized and specific for my obscure taste to have much wider economic value as tools in themselves.
Claude is also the greatest obscure book recommendation engine I have ever found so I have a lot more interesting books to read than in the past.
Outside of this, I basically live in 2019 but everything is more expensive.
I worry this is the future. It doesn't really feel much different other than things getting more and more expensive.
gopher_space · · focus · HN ↗
aduwah · · focus · HN ↗
bartekrutkowski · · focus · HN ↗
tripleee · · focus · HN ↗
ActionHank · · focus · HN ↗
Are you going to trust a vendor who has been running for years with your business critical operations or brad_69's awesome saas skills and agents? Do you think these businesses are going to shift their whole product to be in Altman's walled garden when they don't need to?
weitendorf · · focus · HN ↗
Software engineering on one end is going to be more like finance or medicine where the cost of poor performance is actually higher than the risk that they wont be worth what you pay them (ie there isn’t any amount of pay where it’s worth it to hire them). Then on the other it’ll be like knowing how to use PowerPoint. All the new tech categories have too many bottlenecks and challenges beyond just producing code to consolidate like that.
vlucas · · focus · HN ↗
codingdave · · focus · HN ↗
When they have full DevOps, with multiple environments for testing, compliance with security best practices, can do database updates and data migrations, move the app smoothly between different platforms, streamline costs, and scale to at least hundreds of thousands (if not millions) of users without crashing and burning... then go ahead and worry. We aren't anywhere near that.
On the other hand, if you are purely a coder, and cannot run a project that does all of the above... maybe start your worrying earlier.
win311fwg · · focus · HN ↗
We aren't, but the AI Labs no doubt are. They just can't release it right now because their bread and butter is still developers burning tokens, so there is still a business problem to solve.
rl3 · · focus · HN ↗
That said, getting a frontier model self-improvement harness to work correctly end-to-end isn't trivial.
CuriouslyC · · focus · HN ↗
swader999 · · focus · HN ↗
bharatsuthar · · focus · HN ↗
swader999 · · focus · HN ↗
dofm · · focus · HN ↗
- People in other industries actually have specialist knowledge (this is something IT people forget)
- AI is coming for those jobs too
- and so are lots of other people, not just displaced techies but from any number of other industries.
Thinking about it in advance should include training yourself up for other sectors.
If AI is successful, it will be a very different world of jobhunting. If it’s unsuccessful there will still be job losses for a while because of overspending, and because of the wider economic issues.
dofm · · focus · HN ↗
Sorry, I switched senses in the middle of this.
Should have written either:
"because you should not assume you’re going to walk into another career (I am not suggesting that you are)"
or something like
"because you should assume you’re not going to walk into another career (I am not suggesting that you think you might)"
Need to write my posts less hurriedly.
ryan_n · · focus · HN ↗
dofm · · focus · HN ↗
It's also not clear to me that the level of success in the tech industry is truly being repeated anywhere else. There are industries where generative AI is actually a liability (the law, finance) and where it is increasingly viewed as repugnant (design, the arts, to some extent photography where deep learning noise reduction tools are OK but the culture is swinging hard against other generative uses).
I am also not convinced it is "objectively" successful in the tech industry, considering the extent to which most token pricing is subsidised and people's reluctance to demonstrate ROI even on subsidised/bundled tokens.
I'm not claiming it is unsuccessful. Just that I don't think the job apocalypse is here yet. Though this product (ChatGPT Sites) is very indicative of a coming tech job apocalypse in pure software services.
It's certainly at a level of success where people should start thinking about a "Go bag" on an employment level. Keep your CV up to date, work on gaps in non-tech skills, consider other qualifications.
internet2000 · · focus · HN ↗
Hopefully you've been living below your means and saving.
wingworks · · focus · HN ↗
OtomotO · · focus · HN ↗
If I can't work as a computer scientist anymore (I am my own business since many years), I'll simply do something else.
I have many interests and "talents".
Maybe I can work as a gardener, or I help the farmers around, or I do some social work, or I work as a garbage man, or I ...
I am not worried in the slightest on a personal level.
Society might crumble though.
But then again, was it ever more than barbarism in disguise?
fassssst · · focus · HN ↗
corytheboyd · · focus · HN ↗
On the other hand, death of a livelihood in a social structure that doesn’t care about you.
We’ve always had pump-and-dump trash. The terrible stuff loses and the good stuff wins. I don’t see why that would change.
The companies that ditch the management structures of yore and successfully build better software faster with the new tools will come out on top. I don’t think these companies will replace ALL saas subscriptions with in-house solutions, because one huge benefit of paying someone else to hold that bag is liability and regulatory overhead.
htrp · · focus · HN ↗
weitendorf · · focus · HN ↗
Domains, auth, and databases are probably the stickiest software products of all time. I suspect it’s a combo of security concerns and the risk of eroding goodwill among developers (their most important customer base rn) if they move up the stack too quickly. Once you know that the goal is to become cloud vendors, train on/compete with their own customers, and own the entire software stack e2e it’s hard to really feel grateful that they’re dragging it out but planning on doing it anyway.
For those of us working in infra/SaaS outside these companies it’s pretty clear that the only viable path that doesn’t involve getting cannibalized is training your own agents/models. The new coding agent-infra-data business model is a path towards full commoditization and undifferentiated prompting in 3-4 vertically integrated walled gardens. If you ever start making real money on pure software infra they’ll just be able to eat you alive by implementing something similar, training on their own tools rather than yours, and integrating it into their stack (which all of your customers are on already).
Also, you should only ever use $20-200/mo subscriptions on work you want them to train against or you don’t mind automate yourself out of. Think a little bit about what you’re teaching them to do when you use their products, especially if they’re the primary interface you’re working in. People are going to start caring about this a lot when the AI companies feel safe enough to begin the “extinguish” phase of AI coding. Train your own models!
dennisy · · focus · HN ↗
I do not see enough people discussing this and trying to position their businesses in a way to avoid destruction at the hands of the labs.
weitendorf · · focus · HN ↗
There’s a good chance that the next step after open weight models is something that makes it easier to post-train existing models on new workloads without running an entire pipeline just for yourself. Somebody has to actually build it though.
girvo · · focus · HN ↗
What can most even really do about it? Assuming they pull off what they're saying they will: it won't matter what you do, what vertical, what moat you think you have. It will be a concentration of capital and power we've never seen before, and frankly that's terrible for the world.
weitendorf · · focus · HN ↗
Currently what makes this too difficult for anybody but frontier labs is the lack of access to the full distribution of workloads/data they use to RL multiple separate envs/evals without regressing more than they advance in general capabilities. Because they are the primary buyers/builders of that stuff, they have no incentive or reason to allow anybody to replay or resample it but themselves. And it's all so very expensive to do in aggregate so there's not really demand for other product shapes except more openly available, collaborative/bundled RL that you could plug your specific workloads into (who the frontier labs would obviously not want to support with their business).
However, if someone were to build a collaborative rollout platform that you could use to train private workloads (ie create useful IP that doesn't just become profit for other labs/come from what they already can do), the upfront investment would be amortized over the very large number of potential buyers once it gets into the 5-8 figure range, who essentially have no other choice if they want to remain competitive in the technology industry.
Until openai/anthropic ipo the concentration of capital/spending and ndas/loss of employability is too concentrated for the best researchers to really do this without rocking the boat. And a lot of also-ran ai/saas have the same risk due to the lack of capitalized acquisition opportunities or AI vendors to partner with.
If the plan is ultimately to drive you out of business if you ever build anything profitable with their products, and monetize your knowledge without fairly rewarding or explaining their intent to do so, you might as well defect early and build what you inevitably would need anyway.
you could try to help build that!
Terr_ · · focus · HN ↗
Pessimistically, companies will ignore that and pick the cheapest solution and accept an unknown failure-rate that they assume will be tiny.
Optimistically, they will demand "the AI you just demo'ed, except fix whatever bug so that it never screws up." In that case, SaaS will not be about "training your own LLM model", but about investing in harnesses and rule-engines and GOFAI that constrain the actions of a not-necessarily-custom LLM which is acting as the consumer's UI or concierge. In other words.
pazimzadeh · · focus · HN ↗
Yesterday my dot made a web app to view 3D microscopy files
<a href="https://mucus-atlas-viewer.pazimzadeh.chatgpt.site/?sample=ko2" rel="nofollow">https://mucus-atlas-viewer.pazimzadeh.chatgpt.site/?sample=k...
I didn't even know about the sites feature
ghostghostghost · · focus · HN ↗
pazimzadeh · · focus · HN ↗
On top of it we added bacteria (green) and bacteria sized beads (red). On the bottom are colon lining cells (blue).
The inner mucus layer of the colon was generally thought to be impenetrable to bacteria, serving as a way to protect you from the trillions of bacteria in the gut.
It turns out that actually, it’s really easy for bacteria to swim through mucus by using flagella. We found some bacterial mutants that swim through mucus (the green ones). The red beads serve as a control marking the top of the mucus since they can’t get through the mucus.
Find this and other images in my paper: <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC11784811/" rel="nofollow">https://pmc.ncbi.nlm.nih.gov/articles/PMC11784811/
The real question is then - if they can get through the mucus at any time, why don’t they? And there are various theories about this - which I’m happy to elaborate on this you want. The basic idea is there is a risk whenever you cross the mucus (you get closer to the immune system) and mucus actually feeds and tames bacteria over time by providing them with various sugars and things like that. But if they get desperate/starve, then they will cross.
Mucus is still a mystery - there are substructures within mucus, take a look at this image for example:
<a href="https://mucus-atlas-viewer.pazimzadeh.chatgpt.site/?sample=fimh&view=orthogonal" rel="nofollow">https://mucus-atlas-viewer.pazimzadeh.chatgpt.site/?sample=f...
And here is a good paper too from my collaborator (who actually took these images) who discovered a sub-type of mucus and mucus-producing cells:
<a href="https://pubmed.ncbi.nlm.nih.gov/33859001/" rel="nofollow">https://pubmed.ncbi.nlm.nih.gov/33859001/