I'm being sincere when I say (as I've said on two threads here) that this genre of posts --- "I'm leaving this company I've been very publicly associated with, and here's the new thing I'm doing" --- is deeply cursed. There's no way to say anything interesting without it just stinking like an ad for the new thing.
Obviously, anything at all you say about a commercial project you're working on is easily read as promotional. And you're right, this kind of writing almost always is promotional. But there's a way to do it where at least you're trying to be in conversation with your peers, rather than hitting people over the head with how awesome you think the project is.
But I don't know how to do that in a post like this. I think the only way to read it is as, like, an investor memo. Not my goal, but I don't make the rules.
So my strategy here is just to stay kind of vague, and talk about where I think the world is going, rather than the specific thing we're doing. I can talk your ears off about capability systems, datalog, models driving hardware, virtualization, whatever. Those are fun conversations and I'm very psyched to have them; it's what lights me up about the work we're doing now.
But I don't think it can work here. I didn't submit this post and I didn't upvote it. I wrote it because I didn't want the whole thing I'm leaving Fly.io for to be wrapped up in some dumb Twitter thread.
If you're unsatisfied with the post, I don't blame you, but it's less a bid for the front page of HN than it is an update to my "about me" page. I'd literally rather talk about HN meta, and how to write for HN, than I would about operating systems at this moment. I truly appreciate the interest though.
The only question is if it’ll still make sense in six months, or if an even more unreasonable approach will be the correct one. Feels like a singularity.
I completely agree that we haven’t even gotten past the first post yet in terms of the magnitude of change that is likely to occur. People fail to comprehend that we’re still assimilating the Internet as a means of communication. Cell phones. It will be decades before what we have today is fully utilized, let alone what is likely to be developed. Anyway, thanks for the article.
If you're looking for a UX idea, when the phone boots, it asks "What are the top 5 things you do?". And that's how the phone experience starts.
All the best. (Ignore the silly haters, they're lost in their AI pessimism to just look forward to a new intersting thing).
I'm pretty AI pilled, and I'm building a local-first personal software platform, but I dunno... I've been feeling like i need to cram the "AI spirit" into very specifically shaped boxes to actually build robust software tastefully at speed. Like software on the outside and AI on the inside instead of the other way around.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also.. it is easy to get "duped" by AI without actually shipping. I'm starting to want "KPIs" to help me measure if my time is actually well spent against the things i care about.
I guess I'm extrapolating from your post, but even if the programming language is English, we'll probably still want to spend a large amount of additional compute to produce "artifacts" that have things like performant code, security hardening, and certified taste.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also... when i first got into linux i spent way too long trying to get my desktop perfectly aesthetic because it was all suddenly malleable. Now i vibe more with Debian.
I wish you luck, and I'm looking forward to seeing what you come up with! Fun times ahead for sure.
This is such incredible verbiage. Thank you for saying something so clearly that I've been saying myself in many other ways. I've been using terms such as "providing structure" and "applying constraints" and "reducing degrees of freedom".
I think very specifically the idea of "taste" is going to be extremely important in these AI-saturated days, and it's really where experienced humans will shine...at least for a while.
The model I see is that it is like a butler instead of a home computer, like you are driving into a strange town and looking for food and you are not going to open the McDonald's app but instead your agent does a search and through a conversational recommender you get sent somewhere. It might be sitting in your pocket the whole time and you are talking to it through something like AirPods. Maybe you put on smart glasses and it can show you things.
Apple can't give you the A.I.-native phone because it is too dependent on the app economy. For a few months I've been asking "Who is working on this?" and figured it was one of those cursed ideas that Silicon Valley would be unable to fund much like the pharmaceutical-industrial complex of Cambridge was unable to fund Facebook. I was expecting it to come out of some hungry Chinese manufacturer so I'm happy to see another horse in this race.
(e.g. I love the image that Meta has developed around Muse which makes it look like it is a "butler" when it is really much more like the old Kraftwerk song where "I program my home computer")
I actually think this post gets the point across better than most.
And then of course it's super self-indulgent of me just to write navel-gazing stuff like this. But I'm an HN person first and an entrepreneur second, so here we are.
I am thankful people are being so nice about it though!
As someone who's building its news project on it, it's both reassuring, but also not since you're leaving..
1. When you write a comment in a discussion of an article, be upfront with being the author. It took me a while to understand that from your comment. "Author here" is a thing.
2. Your article itself doesn't have clear authorship stated. At the very bottom there is a copyright notice, but that's pretty generic and not a single person either. Just place your name at the top together with a date. Not everybody knows you or follows your blog.
3. When I read this ...
> When I was a little kid, a family friend sold us our first “computer” — scare quotes because I’m pretty sure it was a VTech Laser 200 clone, a chiclet keyboard Z80 that ran off cassette tapes and plugged into our TV.
.. I prepared to stop reading since .. how is that not a computer? After another paragraph or two which just felt like rambling, I gave up.
You're getting perfectly reasonable feedback here, and your response is weird.
with 3. being the most prominent one.. I made it a little further than that "hot take" but a few more paragraphs in the thoughts expressed seemed neither novel nor to go anywhere. SaaSpocalypse but on localhost... shrugs
These are bad-faith arguments toward "What about self modifying software".
Self modifying software has no grounds in reality or predictable behavior. It is an unreasonable expectation that software is magic. You would not fly on a self-modifying airplane. This is not a future i want to live in.
No; sorry, no.
But it seems that he is falling for the LLM hype the same way. We all have the kid version of ourselves hiding somewhere within our grown versions, it seems...
The author is smarter than falling for the hype, I think. But if not AI psychosis is a hell of a thing.
Depends on domain, but basically all the same lessons we have for software written the old way by humans.
Which, ah, admittedly isn't great.
> Isn't this the halting problem?
No.
1. The halting problem applies specifically to deterministic systems; there may be a non-deterministic equivalent, but not enough people cared before AI got good.
2. For practical purposes, it's fine to reject things that take too much effort to prove correct.
3. "Proven correct" is different from "proven to halt eventually". I guess Gödel's incompleteness theorems would be a partial fit, but even then the goal here is to reject anything you can't prove, rather than the much harder (impossible) challenge of proving the validity of all possible statements it might come up with.
First, the halting problem is only undecidable on arbitrary programs. Here we would be writing programs specifically to be easy to prove correct. In fact, we would be developing programs and proof at the same time.
Second, determinism is a nice property, but it's independent of whether you can proof software correct.
Have a look at eg randomised quicksort or the 'expected linear time minimum-spanning-tree algorithm' https://en.wikipedia.org/wiki/Expected_linear_time_MST_algor...
Or have a look at HyperLogLog. https://en.wikipedia.org/wiki/HyperLogLog
HyperLogLog only gives you a probabilistic approximation of the true number of distinct items. Here the proof would be about the probability that you are within some acceptable window around the true answer.
Sure question what the latest tool can offer and how to integrate with it... but that's just being an app, or a window manager, or a package manager, or a distribution, that's not challenging what the OS itself is and does.
If your OS doesn't change how a computer allocates resources, then you are talking about higher level, and that can be super valuable too, arguably more valuable, but ... that's a different thing.
PS: also chroot (half a century, just checked for the fun of it), jails, containers, VMs, etc limiting how the allocation can be done do exist and have been used for decades already.
In the context of a microcontroller, is it an RTOS? How is that different from something like Embassy with Rust, where hardware allocation and coordination is handled more at build time and less at runtime? In the context of a something showing up on a screen, is an OS something providing a UI framework and hardware abstraction - something like Win32 or Cocoa? What does that make Chrome?
In 2006 it felt like an answer to the question "what is an OS" was straightforward. In 2026, I'm rather confident that I don't have a single, simple answer. If there were ever a time to re-assess what we thought we knew about computing - now seems like a good time to do it.
LLMs are not OSes and never will be, they will always need some software to first bootstrap the hardware before the software which processes the weighting can begin.
Its like we need Richard Stallman to come and argue the difference between GNU and Linux again!
If we ask this a different way, is there a single self executable file when written to the sector of the disk the hardware looks for can contain the instructions for the hardware and the weights all in one?
I can't thing of any informational theory reason why this can't be the case. The only thing I can think of structuring such far more complex and less worthwhile than just having separate layers. Dealing with this biological/intelligence mess in humans is already hard enough as it is.
The OS bootstraps the hardware and allows the software to run. If your software is contained in the same blob then great, but it still needs the bootstrapping code first.
The bootstrapping/hardware management portion on its own would be considered an OS but the LLM code would not, I think thats the main distinction here.
I hear what you’re saying, but we don’t have a word for that different thing. OS means the kernel. But OS also means the entire distribution - all of macOS, all of iOS, all of windows, Ubuntu, Android, etc.
Until we have a standard word for the ui “shell” that comes out of the box with the OS distribution, people - including me - are gonna call (and think of it) as the OS. Or part of the os. Or something like that.
> we don’t have a word for ... the ... distribution
> Until we have a ... word for the ... shell
distribution/distro and shell are exactly the words we have for these things
But OS sounds more impressive/important, so it's a lot more prevalent.
It's really the unix/linux world that stuck hard with these being separate things.
The idea that an operating system is just the kernel seems wrong.
I love your idea of an OS where users write their own apps as they want. But if they want to connect to third party services or use commercial software, won’t there still have to be a lot of security guarantees? I don’t just mean the depressing “browsers must implement DRM or you can’t watch streaming video.” I mean “how will medical providers trust your device enough to give you your lab results” if you are thinking of majorly changing the mobile device security model.
It’s not a hostile question, I’m genuinely curious about this design tradeoff.
I recently replaced my $HEALTHCARE_APP with the PWA of their website. It was the *exact* same. The app was just a webview (and a pretty good one at that!)
It makes me really question how much “security” these services need when they have websites that can do everything the app can do.
The lion's share of value in a product like Salesforce is the ability to say "we run Salesforce" and have everyone know what you mean and have ready answers to all your problems even if they're not perfectly tailored (or cheap). Same thing with OS really. If natural language is now the most powerful programming language then having shared vocabulary on technical topics is more valuable than ever.
In fact, it looks like the advent of AI has actually accelerated revenue growth for SaaS rather than killed it off. People simply aren't building these custom SaaS-killers as much as commenters on HN would have us believe.
Stripe Economics, The SaaSpocalypse was more like a RenaiSaaS: https://www.stripeeconomics.com/p/the-saaspocalypse-was-more...
Streaming is fair. Netflix can dictate their rules. I can decide not to pay for their content.
While I agree in spirit: after a recent rib injury I wasn't allowed to take a picture of my own X-Ray using my phone or even have a copy sent to me since "that isn't the policy"... very hard to say that data was "mine" until it is "released" to me.
There is an infinitesimally small liability that some "other patient's" data might be visible on scanning device, a horde of companies assuring asses are covered and without a legal obligation to give me all "my" health data per visit I don't see an easy path to using our own devices soon, unfortunately.
If you want that power, you’ll need to get your national government to change the law.
I’d love for this to be the case, but we’re going to need laws to make it happen. Banks will have to be forced into it and probably also shielded from liability in case the user does something dumb.
Maybe we can force them into providing APIs for us while we’re at it so I don’t have to pay Plaid if I want to build an app to track my finances.
You were the exception. Most kids felt awe. We started learning BASIC and creating "dumb games". That's the origin story of most people my age who ended up in this doomed industry.
Yet magazines and books commonly published program listings for games in BASIC which many kids somehow managed to type in and run.
>Most kids felt awe.
then
>which many kids somehow managed to type in and run.
The number of kids that had computers that typed in BASIC programs was absolutely tiny, unless they were doing it as part of a school as a grade. The number of kids at home that had computers in around 93 was less than 1/5th of all households and those tended to be the most wealthy. Even then they were so expensive most parents were telling them to stay away and not letting them edit random files.
Until like a year ago.
I would start with a completely blank slate and then, gradually, be able to create something out of nothing. If I was clever enough, it could do whatever I wanted (as long as my wants were modest). It was magical and I have nothing but fond memories of the time I spent doing that.
Trying to correctly render a vertical blinking cursor on top of proportional fonts took me ages to get right...ish.
They can't have sucked that much if you saw fit to build a career around them.
https://www.computinghistory.org.uk/det/1421/sharp-pc-1246/
This was the very opposite of unsatisfying. I don’t recognise this idea that computers sucked.
AI-assisted coding is, to me, way more soul-destroying than hitting the limits of a simple home computer. In most regards I feel like I have reached the end of my interest in computing, and the pivot I had in mind — teaching - looks like it has been joy-deadened too. Nobody is going to want to learn how anything that is not AI actually works, tech people insist that AI will replace (they say augment, they mean replace) teachers and trainers, politicians seem to be listening, and non-tech people now increasingly resent technology.
I barely recognise the industry I spent my life in; I do not recognise the evident enthusiasm for a world that is abandoning a whole, transformative pedagogy in favour of the programming equivalent of games consoles: button pushing vibes.
I am not at all of the opinion that AI has no value, though it does not have anywhere close to the value people insist it must have for me. I am learning, trying to find an ethical angle on it that doesn’t make me feel uncomfortable and exhausted by the hype.
But it is the final nail in the coffin of the joy of tech work for me, and as a single middle-aged man I perceive this as life changing and very likely life-shortening, in the way that psychologically unhealthy things are a health risk for older adults: unless I find something that gives my work life a sense of connected purpose again, I figure I will be gone before the Spectrum makes it to fifty.
So yeah, no. No.
I disagree with most of your post, but this one the most.
Non-tech people already resent technology... Try and sit next to a non tech person use a desktop pc. It wasn't until the smart phone that the average person started to use technology like you or I recognise. And that's only because it's been simplified to the extreme for them. (I don't blame them btw).
AI is such a step change for learning and teaching... For anyone who is inquisitive. Ask it endless questions, explore ideas, get it to generate images to help you...
Same here. I can understand indifference, but this I don't get.
FFS I even loved playing with calculators!
Why did you think that? Did you see people use it like that? anything? really?
For me, computers became intriguing after I started programming in BASIC. Not before that. Before that I was just curious about what the big deal about these things are...
In the literature, there would be passages that describe how computer followed "instructions", but I never could make sense of that or understood what an "instruction" was. So it didn't really sway me either way. I just tried hard and remained curious to understand what this "instruction" was and how I could give it to computers...
May be your falling the same way for LLMs now..
I have never found computers disappointing, and I have spent most of my career/life learning about them - from processor design through to distributed systems. Mostly I've been disappointed in the disparity between the number of interesting things I'd like to do, and the number of lifetimes I have available to me.
Different strokes for different folks, I guess.
Yes, but they would say the same about any other programming language too.
BASIC was common on the Apple ][ series, and in the Win95 era QBasic was popular.
It was not impossible to have access to computers in the educational environment (either in class or as an extracurricular) before middle school, either.
20 goto 10
20 goto 10
The vast majority of kids never touched computers, other than maybe running logo at school to move a turtle around.
Most kids who were only interested in a game had a NES or Sega console.
Of the former group, the awe began when you wrote in the example program from the manual or magazine, made it do something, and then changed it. Even something as simple as changing from 3 lives to 99.
Having been one of the interested kids (but it took quite a while before any real program design or analysis clicked for me) my experience was that most of the others didn't seem all that interested (and were really just there to play games).
Then you got the computer and when it started up you got a flashing K. It was a bit of a disappointment. But you flicked open the hefty manual, started reading and before long you were hooked.
One career later, I now have a computer that is very, very close to what I wanted all those years ago but with a significant and irritating genie attitude to granting my wishes.
Now if we can just tone down the genie stuff…
Some kids thought it was awesome to make the computer do stuff and learned how to program.
BASIC was designed as a simplified, interactive version of FORTRAN that non-experts could learn in an afternoon, but which was still powerful enough for a wide range of applications including numerical methods, simulations, data processing, and games. It was also small enough to implement effectively on a timeshared minicomputer or an 8-bit microcomputer.
Even though the Mac was GUI-based, for a time it shipped with, and promoted, HyperCard, which led to a brief explosion of user-created programs developed as HyperCard stacks.
In addition this also increases the probability that your unaware of others peoples drives, likes, and abilities, which can easily mislead you to believe that a large portion of the is skilled like you in the same way and wants to conquer information.
Yes, but now everyone and their mum has a personal computing device, and there's tutorials readily accessible, and five seconds of googling can bring you to an online programming environment.
As I mentioned, I think graphing calculators might be an exception.
I learned it at age 12. It was pretty much as the author described, except in my case the "computer" was a Texas Instruments teletype, a phone, and a 300 baud modem. Plus a booklet with reddish cover introducing the BASIC language and how to connect to our school account at Washtenaw Computing Services. That was the beginning of a life-long journey with computers.
My Apple //e came with one language, BASIC, it was built in, nothing to buy, and came with a book on how to get started writing it.
But yeah I didn’t start at 8.
We didn’t get that Apple until I was 10, so that’s then I started.
I actually still don't know BASIC.
I was really surprised by the article, I assumed it was going to be the same as my experience which was amazement at having the ability to write instructions that caused the system to do stuff, like make dumb games initially in basic and then better games in assembly.
But 8 is not too early to learn programming as such, because my kids started around that age. Only they started creating dumb games using Scratch in a modern browser on a full-fledged modern OS. The environment you start with probably matters a lot for how much you take to it.
But I think we were the exception, even if we didn’t realise it at the time. In hindsight, my interest was most definitely not shared by my peers.
I learned C, Java, JavaScript and Obj-C from books over the years. Those books were honestly much better resources for a beginner programmer than most online resources are today.
QBasic's bultin help/documentation was good enough for me..
As I mentioned in another comment, I think if you didn’t grow up with it, you just underestimate how amazing QBASIC was as an IDE, a programming language, and perhaps most importantly as an educational tool. I mean, it even came with actual BASIC programs (GORILLA, NIBBLES, etc) right there that you could use to explore and tweak the values as you learned.
There’s no better way to learn how something works than to physically play with the actual values, change constants, change variables, and observe how that affects the behavior of the game.
If I remember correctly, there were even sample programs included in this documentation.
For example GET/PUT had this one:
SCREEN 1
DIM Box%(1 TO 200)
x1% = 0: x2% = 10: y1% = 0: y2% = 10
LINE (x1%, y1%)-(x2%, y2%), 2, BF
GET (x1%, y1%)-(x2%, y2%), Box%
DO
PUT (x1%, y1%), Box%, XOR
x1% = RND * 300
y1% = RND * 180
PUT (x1%, y1%), Box%
LOOP WHILE INKEY$ = ""I hope this makes you feel better: all industries are doomed. But few rise and fall in one lifetime.
..at that time I have to admit, most of what I saw there went over my head. I understood the concept of program code, but had very little idea of what was going on, but wanted to learn. Eventually I started modifying the code, and started making my own stuff, driven mostly by determination rather than any understanding of what was going on.
There’s a huge difference between a kid turning on a Commodore 64 or an Apple II and being dropped into the relatively no-frills versions the BASIC interpreter versus a kid in the early ’90s who had a DOS machine with MS-DOS 5 bundled with QBasic which IMHO is a more representative example of a “batteries included” BASIC environment that was significantly more user-friendly.
I also encountered QBASIC as an elementary school kid - it felt like the closest thing to sorcery. People who grew up in the age of the internet really underestimate how awesome it was to be able to just be able to place the cursor on a more obscure command (like PEEK/POKE or POINT) and see something like this:
Returns the current graphics cursor coordinates or the color attribute of a specified pixel.
'This example requires a color graphics adapter.
SCREEN 1
LINE (0, 0)-(100, 100), 2
LOCATE 14, 1
FOR y% = 1 TO 10
FOR x% = 1 TO 10
PRINT POINT(x%, y%);
NEXT x%
PRINT
NEXT y%Then I transitioned to Windows, and had to contend with all the complexity.. What's a Window Class and how do you write an event loop to respond to window messages, why is this all so complex.. Why is MFC somehow not really better.
Having written a simple DOS GUI (that kinda looked like Windows), I had to discover that a Windows app of equivalent functionality took more and messier code.
And forget drawing to the screen, you had to grapple with GDI drawing contexts, it was slow and bad.. (Not like Linux was any better than this regard).
As for DOS, I have had the privilege of hacking together a bitbanging VGA adapter, so I know for a fact that drawing to the screen is really as simple as setting values in an array, then, scanning it out to the display.
Screen mode 13h was definitely fantastic. Being able to light up a pixel by mapping directly to the 64k contiguous bytes in the address space was very intuitive.
And yeah dealing with Win32 stuff definitely came with its own set of complexities, I practically had the Charles Petzold Programming Windows book memorized back in the day.
But let’s also not forget Visual Basic, which outside of maybe Delphi, I'd consider one of the greatest RAD environments ever created. Being able to just grab a button onto a form and then double-click it which would instantly create a handler that you could add your logic to was very approachable even for younger audiences.
The only problem of course is that Visual Basic wasn’t bundled with Windows, so it was a lot less accessible than QBasic which came bundled for free with MS DOS 5.0.
Today, computers don't even have parallal/serial ports. Everyone has to pay the USB gods their tribute to have the mercy of allowing to program arbitrary IO from their computers....And they still need some kind of serial receptacle on the other end...
OPEN "COM1:2400,N,8,1,BIN" FOR OUTPUT AS #1
“Mom, I’m bored will you buy this a new computer game?” “Lol no! Go outside.” “Well crap. I guess I’ll have to write my own game.”
Adopters feel the future is brighter than ever.
Article author is excited. I am too.
I can give you requirements, give you feedback during the process, and otherwise contribute to the design, but at the end of the day I did not build the house.
I don't write assembly, but I still had to learn how to code. I had to practice and put in the effort to develop it as a skill. Then, when I write something, I have to put in the effort to do it.
Of course, building things with AI generally takes varying levels of domain knowledge and guiding for now, but if we're talking about ordinary users making their own software, the implication is that in some cases existing skill won't be necessary, if not now then eventually.
At that point it would be truly effortless, and I don't think that you can rightfully say you built something if it required zero effort on your part.
(Raises hand.) I'm doing things with my online law-school course materials, in just a few minutes using Claude, that I never even considered doing before because I didn't want to take the time to improve my rudimentary JavaScript- and CSS skills. Instead of just a fairly-simple, single-page HTML document, Claude is helping me turn the materials into something genuinely useful for both my students and practitioners.
One of my favorite journalists is Jerusalem Demsas, who just on a podcast mentioned --- in passing --- that she'd vibe-coded the photo editor they were using for her publication. That's a bananas statement! And we're all just, shrug, or worse: "that's slop!"
Don't get me wrong, I think we'll print money hats off this sentiment. By all means, do whatever you can to leave the lane we're building in as open and stigmatized as possible!
It also sucks that so many people in tech think that the widespread loss of education, knowledge and freedom is a good thing.
I get that it might suck for people like us, who took the time to learn those languages! Oh well.
No one is saying that anyone has to, but being able to is the whole thesis of the hacker ethos. Knowing how to do things yourself versus being a consumer dependent on corporations.
And children are learning how to program nowadays... although that won't be the case for long. My nephew is taking programming classes in school and making games in Roblox. But soon the only thing they'll teach is how to prompt Claude. We were almost at a point where programming could become general knowledge and that might have enabled widespread freedom and control over the evils of proprietary software and commercialism. Now we're burning it all down and a depressing number of people like yourself on this erstwhile forum of programmers and hackers are actually dancing on the ashes.
Using a service to generate software for you is not "building programs to solve your problems." You aren't building anything, and you certainly aren't learning anything that would give you any real control over it or let you exercise freedom relative to it. AI is just proprietary software and data theft with more steps and a lot more fascism.
It doesn't just suck for us, it sucks for everyone.
Why shouldn't one lament sacrificing individuality and creativity, and increasing our reliance on corporations, all for the sake of convenience?
Ease and convenience are not worth the cost of giving up bits and pieces of our humanity here and there until there's nothing left; Pre-AI computing has already taken enough of it.
> I do not think it's a good thing for normal people to have to learn programming languages to built programs to solve their problems.
Is this not also "orienting all of society around your aesthetic preference"?
Lol. Not only was this not true but we weren't even close. Genuinely bizarre that you could think this.
>enabled widespread freedom and control over the evils of proprietary software and commercialism.
Ironically, the only reality where this is remotely feasible is one where frontier llms can be run locally on phone like hardware and it becomes popular to do so.
You can and should still write most of the code yourself and use AI to assist when you get stuck. Apart from saving yourself some hours because you no longer hit dead ends on google, I don't understand how that's meaningfully different from using a search engine and finding solutions on StackOverflow and GitHub. I don't even understand how people think this traditional workflow I'm suggesting is any slower than reviewing LLM output either. It takes the same cognitive load and the same time in the end. The only people who believe they can go faster and reduce cognitive load aren't just inexperienced, but also seem to have ADHD or something. Like they just wanna bask in the vibes of the machine-generated text scrolling by.
If your goal is to make something you're going to live with and maintain long term, or is more generally the foundation of a bigger idea, you can't ignore any part of the code or delegate it away. It's still writing that you're bound by like any other document.
The "strawman" comes from the businesses promoting the frontier LLMs and many of the best-known influencers in the space, especially on Xitter.
Are you telling us the only reason why making things was fun is because it made you "special"? And now that anyone can do it it isn't worth it for you anymore?
That's the wrong reason to do something.
Unless you're complaining you can't be paid to do it the old way. But none of us were paid to have fun, we were paid to deliver results.
I am not entitled for things to continue as they were, but I can still feel sad about it, you know.
I feel like the future is bright precisely because AI sucks so bad while also lowering the bar.
It made people not take software and software engineers for granted. Public perception of software has also finally matured. Boomers and tech CEOs fell into a tarpit that harshly exposes their stupidity while laypeople now feel like they can change the system.
This is all exciting in the best ways. My usage of AI at work and for personal projects goes as far as whatever the AI summary is on google. That's more than enough for me to get enough of a productivity boost on par with those who embarrassingly thrust themselves balls deep and refuse to pull out. Not everyone has to push the "I'm feeling lucky" button to get the same work done at the same pace.
I guess that means you are still interested in pretty much everything
Please.
I am at a job where I am paid well to build software to do things in a distributed system that processes information across various external systems (some with physical world impacts). The outcome of that computing is revenue for the company. Using agent-wrapped LLMs makes the software authorship component of the job really fast. The rest is virtually the same speed.
For Thomas's mobile computing examples, these are done really well by my phone's operating system's built-in voice command system, mature for a decade by now.
At home, I use software that is well-crafted to do some thing. Or I write a series of scripts to do narrow things that for some reason aren't in Aptitude or Snap, for example "copy all the photos off my iPhone and rename them and sort them by date in my filesystem".
Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
Thomas has been a successful cryptographer, vulnerability analyst, and entrepreneur, so he probably knows what he's doing. I just don't get it.
> Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
I do not use LLM either, and do not intend to do. I can write my own programs. Even if someone does want to use LLM, it should neither be required nor expected nor the default (and there are also reasons why I think that it should be discouraged to use LLM too much); you should be able to write the scripts and programs by yourself if you want to do it that way (and the computer made to make this possible to do without LLM), even if some people can make LLM to do it at their option, but it should not to require LLM and AI.
I also think that the computer that can start with a programming language such as BASIC or Forth (without first needing menus and GUI to access all of the programs) is useful, and that it is not as good that many modern computers do not do like that.
It feels to me rather analogous to being in a foreign country and, rather than learning the local language, you get to hire this translator that works for absurdly cheap, but you have no real idea of the accuracy (because you didn't learn!) and for some reason the translator is only able to speak while physically holding your wallet.
Oddly enough, from what I can tell, AI-powered translation services are not capturing a lot of imagination, while interest in AI-powered (human) language tutoring is booming (at least on the supply side).
Having chat gpt one write me a powershell script took a 3 hour manual task using software and made it a 10 minute task which I can repeat any time I fancy should I need to.
Writing something similar would've taken me an hour+ just to dig out the key pieces of information I needed to start in a language I am proficient in. Why would I waste my time?
Nothing. Thomas, as most ambitious start up founders do according to PG, is living in the future.
There are already a few places that are living in the future in this respect. Silicon Valley is one of them: an unknown person working on a strange-sounding idea won't automatically be dismissed the way they would back home ...
The right way to deal with new ideas is to treat them as a challenge to your imagination — not just to have lower standards, but to switch polarity entirely, from listing the reasons an idea won't work to trying to think of ways it could.
Unless you have a local model and the appropriate hardware for it, your agent is Somebody Else's Computer. Do you really want to send all your data and make your entire computing experience dependent on whatever OpenAI or Anthropic or whoever else is planning in this moment?
That's not even starting with inference time and token cost. Despite all the incredible advances in inference, it still takes more time than most non-AI computer functionality. Do you really want to wait a few minutes and pay money for something that you could also do with a few clicks fully locally on your PC?
But the most important thing: User interfaces. Right now, we're basically cramming everything you could possibly want to do at a computer into a chat interface. But there are lots of applications fir which specialized graphical interfaces are much more suitable. Why would you want to get rid of them?
Somewhat connected to that: Repetition. If you have to do the same task again every week or every day, it seems wasteful to ask the AI for it every time: Not just are you wasting a lot of time, energy and tokens, you're also at risk of getting inconsistent results, if the agent from today's session will interpret the requirements slightly differently than the agent from yesterday.
You can circumvent all those things by having the AI write you a custom app, but then using the app without AI to do the task.
I'm not a Treky but I think Star Trek has the right interface here. For basic operations you just use voice and tell it your intent or what you need to know "Computer, what's the fastest route for me to get home?". "Madison is fastest right now, it's 3 minutes faster than Lake, it traffic is building up".
This isn't just whatever Waze gives you, it knows about your constraints and preferences. So you need to program that somehow. Imagine taking it one step further "Computer, what's the fastest route to pickup my daughter and get home?" or even it just responding to the first prompt "Take Roosevelt, and today you need to pickup your daughter from volleyball".
In Star Trek they often do this when they create "simulations". They tell the computer constraints and then it refines it. No writing programs, but effectively that's what you are doing but with the natural language voice interface.
"Computer let's create a simulation to figure out the fastest way for me to get home. I want to take a route along the expressway and know whether to get off at Roosevelt, Madison, or Lake. I need to pickup my daughter when she has volleyball." You don't need to repeat the whole instruction each time, and behind the scenes it is probably turning it into some deterministic code, so next time you trigger that program it's not going to hallucinate some nonsense.
My point is that text-based interfaces - and voice is just a convenience around text - are not the best interfaces for a lot of tasks.
Imagine saying that to a random dude with a map and that should give you a good idea on how many info you're leaving out. You can say that the AI will infere those but that seems to be the common fallacy of AI enjoyers: They're always assuming the AI will magically find out the missing information from their prompt somehow.
I have Organic Maps installed on my phone and within a few minutes (faster than my car warming up) I can set a multi stop route (you can bookmark places). Training a random user to use such apps is also equally fast. The same happens with various pro tools: once trained, a user can be very fast with them.
Wow that's slow. I can do the same in Google Maps in seconds. Can you tell me what I'm missing? I'm definitely missing something.
> just drive easy for a couple minutes.
Fortunately our industry has been getting people used to these for years.
Sometimes it is wild to me how two people can look at the same information and come to completely different conclusions.
As best as I can tell, the only constraint the really matters right now is availability of compute, or as a buddy of mine put it "GPU GO BRRRRRRRRRRR!"
So, any Apple computer?
Going to entirely depend on the nature of the task and where AI pricing goes in the future.
Would you like to spend $0.10 on every task or would you like to spend $1.00 to make an app that does it free for you forever?
I don't think it'll be one or the other.
But also I don't see things like video games ever being replaced by 1 off AI. If for no other reason then the fact that a lot of games are fun because of multiplayer capabilities. That's not something you can do a 1 and done with AI (unless you only want to play games with family and close friends).
With the caveat that literal today's models are not allowed to run code themselves but instead give you the code in the chat window and ask you to run it yourself for security and safety reasons. But the capability is there.
A tool for generating a report for September 2026 is no more expensive than an agent generating a report for September 2026. But that tool is useless for you in October 2026. You'll need to generate a whole new tool for October 2026, and generating 2 tools costs as much as generating 2 reports so you're not saving anything.
Making a tool that can generate a report for any single month can easily cost 10 times more than a single report, entirely due to the extra code flow analysis required to extract the right variables in the right places. And that's without any UI. If you want UI that's another several times more tokens because UI is hard. Also it will suck compared to a simple chat window where you can ask the assistant for anything and it will understand. If you want UI that's better than the chat, that alone costs you more than everything you spent so far.
All in all, your permanent solution starts paying for itself in about 4 to 6 years, assuming you never had to update how the report is generated in the meantime.
Or you could have a very simple, very cheap script that only gets run once then gets deleted (or even better, kept for future reference to make future report scripts even cheaper to generate).
Note that this is all in the hypothetical future where AI is actually capable of one shotting such a task. If it makes mistakes, then human-made apps are still the most cost-effective option and this whole discussion is moot.
Applications are compressed bundles of some assumptions about how something should work and be interacted with. So you wouldn’t need VLC, it’s a shell around a foundational codec layer. Instagram is just a view over data in the Instagram social network, and you could customise it as you see fit. Transmit is a lovely application but you can delegate that to a fluid task layer that can do one-off tasks or make something you can re-use.
Everything in the fat ‘middle layer’ of software that is connecting a user to something foundational through interface assumptions might disappear, unless it’s more useful for it to exist as a permanent tool. Things that a lot of people share or that are more complex, Blender for example.
I also imagine there’d be a lot more focus on making explicit things like canonical records or schema, contracts, provenance, versioning, access policies, some kind of fixed views. Things that your local fluid interface would rely on the structure for when it came to connecting to your bank and doing whatever you want with your finances.
Not all software as we think of it might disappear, but the things that are lightweight layers over something more stable could all be subsumed with tools synthesised on demand.
Perhaps, at least.
Sometimes you really want the latter even if the former is less effort.
So it would really be beneficial for the LLMs to be able to have their own environments to write and run code, but also to store those programs for later use automatically.
A lot of our manufacturing ability improvement during the last 200 years, as well as our engineering ability improvement during the last 500 years, is not just the raw power we unlocked, but that we made it deterministic, reliable and reproducible and hence automatable and automatic.
Can you build a new hospital? Up to the local government to decide. How much is your property tax? Only a tax assessor can tell you. Which trees get trimmed and which are left intact? Decided on case by case basis with no strict rules. Which convenience stores are getting audited for tax and food safety compliance this week? There's a middle rank officer who makes that decision every week, my mum used to be one. How high is the bail set in criminal proceedings? Judges are famously very inconsistent on that and it's a good thing. In countries without constitutional gun rights, it's common that the final step in the process of obtaining gun permit is some government employee having a good gut feeling about you, and if they get bad feeling you get denied. And so on and so forth everywhere you look.
Manufacturing is a different beast entirely. I'm not saying we should put Siri and Cortana in charge of bolting down Boeing fuselage. But for most office and bureaucratic tasks, demanding perfect determinism is not only unnecessary but actively harmful.
Nondeterminism is actually a bug in many cases, and training, process and procedure, is how we mitigate it in humans.
Defining every single step of every single task and making them perfectly repeatable works in an assembly line. But as an office assistant - and doubly so as a personal assistant - it creates inefficiency, makes dealing with edge cases very painful, and greatly increases maintenance burden of keeping all procedures up to date. Instead, giving people some leeway in how they approach their tasks and deliberately underspecifying processes to let them exercise their judgement is shown to work much better. Surely you had it covered in your training materials?
Computers can do that!
That said, my point is that determinism is incidental to how current generation of compression algorithms work, not an explicitly desired quality. And even that isn't strictly true, because e.g. x264 in certain multithreaded configurations already produces non-deterministic results that look as good as single-threaded encoding. For this use case - and thousands of others - losing determinism when switching to AI-based solutions really isn't a big deal.
The LLMs empower for building, the determinism improves output-expectation alignment. It’s fundamentally an on device software factory, hypervisor, and storage system.
Remember when everyone was talking about treating computers as cattle rather than pets?
This seems like a weird inversion. Programs as pets instead of cattle. Toxoplasmosis, ahoy...
Think of all those bad sci-fi movies where aliens take over Earth, enslave all humans and treat them like cattle. Very competent cattle that's put in charge of local governments, scientific research etc. but very much still cattle where individuals are entirely interchangeable and killing them is just everyday work. This is me and AI assistants. (I'm the aliens.)
Pros and cons to code being cheap to generate now, it’s always about tradeoffs. I don’t want to build cars, I want to drive cars.
People suck at "good enough". All major engineering disasters were something "good enough" for somebody. People burn down their houses with "good enough" electrical work.
I doubt it. Most people would rather stab at their phone to download an app that just does what they want than explain to a chatbot over a period of minutes/hours and several reiterations to get it to spew out what they want, let alone what kind of software they want it to make for them. Most people can barely articulate what they want in the first place.
The market for applications isn't going anywhere and neither is the OS. I can't imagine booting a computer directly into a chat window where I have to explain to a bot what I wish would show up on my screen, where on my screen, and how it should look. What a total nightmare that would be!
Even if what passes for "AI" these days was as smart and capable as the computers on Star Trek (and it very much isn't) you wouldn't want to ask it for everything. Even the Enterprise had an OS that people interacted with. What they also had though was an OS with configurable interfaces that people could customize according to their current task and preferred workflow.
That sort of thing is maybe possible in the future for "AI" empowered operating systems we might live to see, but 90% of the people using computers aren't going to bother with any of that, just like they've never seen the settings/options in most of their applications or their operating system right now. Not because it couldn't make their time working the computer better, but because they just can't be bothered.
The writer of this piece remembers how, as a child, computers promised to let him write his own software and games and it seemed like there were endless possibilities until the reality that programing takes effort killed a little of that enthusiasm (although not enough that he didn't eventually become a programmer), but many people never see a computer as anything more than a means to an end or a device for mainly passive content consumption.
Most of those websites have been complements for Google Search rather than standalone services whether they like it or not, and if chat bots replace Google Search, they'll replace the entire ecosystem.
Not that this matters, because Google search alone disproves the claim that people only use phones for videos and social media.
I do personally think that we'll see regular users interested in new, different workflows than the automation ones we associate with power users, and the image generation use case you mention is a good example. Chat-bots do seem to be owning those rather than entirely new apps.
Define "soon".
Banks in the UK have thirty million customers each. They have moved away from web interfaces to apps on well-controlled platforms.
Social media doesn't use the mobile web; it's not good enough.
Neither of these applications for a phone are simply going to surrender to agentic environments on a phone that de-emphasises apps.
They both have a need to fully control that experience, and yet they are polar opposites in terms of seriousness.
We can't even stop voice assistants acting on commands they mishear from TV shows, youtube clips, radio.
“Go play video games for me”
“Go create music for me”
Idk. Seems like there is still use for apps.
Fast forward a few months, and most of those apps are abandonware. Not that he's not as productive, he just skipped the middlemen (the apps), and started directly asking the agent to do those things.
Given the current trends in predictive language models, once users are able to articulate their needs with clarity and precision, conventional software development may become obsolete.
Current capabilities are not there. You need software engineers to keep it going. And even then as intoxicating as it might be to be able to bring to life all the side projects you often had no energy/skill to finish, today it is still may be cost / “cognitive load” -effective to buy off the shelf apps that somebody else has to maintain (even if they are not as personal / less customized for your use-case).
Apps will become barely visible data providers. Some will provide specialized widgets for those cases where a specialized UI is needed. And data will be designed to inter operate.
I have reached this conclusion from a specific example I'm working on, more details in this blog post if you pardon a link to a website of mine: https://cloudywithachanceoflatency.net/blog_02-how-llm-accel...
I don't understand why people are still thinking that clicking on stuff will still be a desirable thing in the future.
What happens after an input is still the best part of a UI. We like seeing/hearing/feeling things in different ways.
I don't want to talk to my computer to use it.
LLMs are the first machine learning models that blatantly and regularly produce incorrect output and we have been brainwashed into accepting that. An application on the other hand can be exhaustively verified. There is no comparison.
If we don't assume up front that AIs are good enough for at least one of those things, then there's no conversation to be had. Personally I'd rather have a conversation than not have a conversation, but you do you.
To me they're failing to see the even bigger picture that maybe even Hey itself is outdated in the Ai era and not only is its Ruby framework antiquated... the whole Hey service is too.
I suppose if you break your glasses, squirt real hard, and drop down about 30 IQ points, then sure.
We're extremely far from anyone, even those with the most interest and capability, giving a practical demo of this future you're suggesting.
What we all see irrefutably happening is some new people are just barely able to do slightly more than they used to. That's the impact we'd expect from a better search engine and a better "I'm feeling lucky" button which, of course, is what this era of AI is.
Yes, absolutely. This is a very far future. But that future is (maybe, possibly, hopefully) coming, unlike this other future which is absolutely never coming.
Having AI "do the task directly" sounds just as miserable as anything else people keep insisting is the future. We can't rely on probabilistic methods. We will still need humans in the loop to pin down the deterministic parts and also the parts we intended (often the same, but not always).
What else did you have in mind?
If AI keeps getting better, then it is inevitable that voice conversations will become the dominant way in which mostbpeople use most computers for most tasks most of the time. It's just part of human nature and there's no way around it. The only way this doesn't happen is if AI never becomes good enough to support it.
Personally, I'm not a fan of voice input. But I want AI to get better. And the future where voice conversations work well enough to be used as the main HCI method is inevitably also the future where voice conversations are used as the main HCI method by most people. I'm hopeful for technological progress, that's all. It's just that you cannot have this progress but not have people take advantage of it in this exact way.
Compare it to the future where most people vibe code their own apps, which has zero chance of happening no matter how good AI gets.
"ChatGPT, please send <username> a message with the following contents: <content>. Oh, what platform? Create the platform to send the message for me. That username doesn't mean anything on this brand new platform? Uhhhh" [insert step where the AI-psychotic decides to give up on communicating with humans and tells ChatGPT to roleplay as the person they intended to message, I guess]
In other news, switching from gas guzzlers to electric cars had the unfortunate side effect of highways disappearing.
Sounds more realistic than what this blog post is proposing. Meta also tried releasing a phone, how did that go?
Muse on the other hand is the polished version of what OpenClaw should have been and Meta appears to have given a lot of thought about it more than this blog post.
I don't think we need yet another phone.
If I'm running software written by an LLM, even if I was the one who prompted the LLM, I would still want my OS to treat it as if written by a stranger.
I know he's not necessarily talking about relaxing security models here, but he's not being very specific about his vision otherwise.
'Transacting' requires not only identity but accountability, a completely controlled-by-you phone cannot meaningfully transact separately from you.
I feel like a lot of new security concerns would be created.
Problem with sandboxes is that it doesn't solve the "user who doesn't know or care" problem.
If you have to give access to particular files to an app, your average user will just give access to all. Someone installing tiktok doesn't look at the permissions...
The best thing would be for apps to operate on their own copy of a file regardless. But that doesn't stop bad apps from leaking importsnt files. Unless files can be signed as being for x thing requiring y special permissions as granted by the originator and not the user.
The crowd on here we can't to be able to truly own our devices. But I think for the safety of the average person they should be locked down by default, same as dev mode on android devices where you accept liability for doing silly things.
(Yes reworking our all our systems in terms of capability security is a giant job. But there's a difference between a giant job and )
I didn't submit this! Happy to chat about it, but I feel like I was pretty clear in the post that I wasn't bidding for the front page.
I've been feeling this way for months and months. If you're a person who's curious about computers and what they can do you can do for you, we just kind of talk to the computer now and ask it to do the thing. And it does it. I know most of what's happening the whole time yet it still feels like magic. We're pondering orbs all day now.
I've been a professional software person for 15+ years... my assumptions as to what the right decisions are in building software now are racked every day. And more than ever I want the device you're describing: blank slate hardware I build to do what I want how I want it to. Yes, I know that's not for everybody. But I really want to just boot up something from scratch and say, "this is how we should deal with email." Because the way I want to deal with email is the way I want to deal with email, and LLMs will help me get there. Get all this UI chrome out of my face.
Last, I've also mused (heh) about a sort of GitHub for PRDs:
> ...there’ll be thousands of times more applications that pull in 1/7th of the guts of a word processor to solve some idiosyncratic work or home life problem for somebody who doesn’t know what a for-loop is.
I think those guts-as-md-files can be maintained in on publicly available platform for all to use. Sort of a "best practices of the commons." Language-agnostic. Or rather, one language: English. Apps going forward don't necessarily need a datatable plugin — because every datatable plugin eventually has defaults or assumptions that clash with what you need — but it needs the schematics for the types of parts a datatable can have.
It's just such an interesting time to be curious about computers.
All you need to do is look at history to realise that. We had things like BASIC, HyperCard, and the early web that attracted droves of people who were excited by the opportunities. While a lot of those people went into industry to write software for other people, a lot did not. Most people, I suspect, didn't even explore the potential because they simply don't think of computers as programmable machines -- even if they were aware that there are people who program computers for a living.
The barrier to entry is lower now because of AI. Even though I code, I ask the AI to write a lot of little helper apps for my wife who homeschools our kids, and I don't code review many of them. I only do enough user testing to ensure they function as requested.
Armed with this knowledge, I have tried in vain to get various family and friends who are not technically minded to do anything like this. In some cases, they just tell me that they can't think of anything the AI can help with.
Many things that seem simple to them are more complex, and they can't tell the difference between simple apps that the AI can knock out in a few minutes and larger projects that require internet hosting and/or vastly more AI credits than they would ever be willing to pay for.
By far the biggest problem is that many of them are not very good at describing what they want. For simple apps, you can go back and forth with the AI to ask for changes, but the more complex the solution, the more its becomes like a traditional engineering design project where carefully though out design of features and their interactions are critical for success. The new "code" may be more like human language, but the user facing details must still be specified.
Claude especially, where you might say “how can I do task xyz more quickly” and Claude responds with a little JS app. This creates an expectation that it’s possible to do anything if you ask Claude right.
I have seen this first hand where my partner - a deeply non technical person - came to ask what an API token is and how she can get one for Canva. Claude was busy automating her Canva -> Pinterest workflow.
She now has a half dozen of these things and hunts through her Claude chats to reuse them.
LLMs are simply the latest incarnation of VB6 + Clippy.
In my opinion, an OS that hosts a local agent can produce things for you on the fly in response to a particular task in addition to the baseline set of features.
Too often have I ran into "you can't do that" in operating systems, especially Macos where you either have to buy a paid app to implement what they should have already, or what you want is niche enough that you have to build it yourself.
Plus and agent can operate across your hardware including the cloud. I could ask something like "graphics cards are expensive and always out of stock so when some pop up can you notify me and show me the price etc" and then receive a rich alert on whatever device I happen to be using at the time it happens, once agents are aware of where we are and what we're doing (yes, it will be pretty dystopian with current governments and the drooling animals who vote for tribalism).
Or being able to be like "I want my photo gallery to show animated thumbnails when I've taken a series of photos of the same subject. Oh, little bit slower like one every second thanks" boom done.
And much UI will becomes defunct in the first place, UI only exists because of many annoying discrete steps we have to do to get something done. If I can just ask an agent to do the thing I want then I don't care about that.
It's like how I only need to make a PowerPoint now and then and every time I open it I've completely forgotten where all of the stupid fucking buttons are in the stupid fucking ribbon bullshit they made.
I agree with BASIC taking effort, yet the baseline for effort has also changed. Over the span of my lifetime, we have gone from virtually everyone performing tasks without the aid of computers, to a few hobbyists and early adopters writing their own software, to buying software off the shelf, to using targeted online applications for free or nominal cost. At the same time people are expecting applications to be easier to use. As you suggested, constant churn in infrequently used applications can cause a lot of frustration.
Now I agree with you if the discussion is about using AI to immediately solve the problem at hand. I simply don't agree that it will end up with people creating reusable applications to perform the task. While overcoming the technical hurdles would be easy enough, I think that it takes a certain type of personality to think of creating a long term solution. Then there is describing the problem in a way that an LLM can translate into code, then there is the testing of the software generated.
Looking towards macOS and iOS for example, an underrated feature for the “it just runs your web browser” crowd is all of the system-provided databases. These turn rather specialized systems - photo libraries, music streaming, calendaring, contacts management, and mail into shared databases that link applications together.
On iOS it goes much further - because of HealthKit, my Withings scale, my Garmin (when I had one), my MyNetDiary nutrition tracker, my continuous Glucose Monitor, and my third-party analytics apps share the same access to my data. And it is done securely - I can choose exactly which part each sees. This is a tremendous resource for both developers and LLMs.
My calendar, hosted on a personal server but available through Calendar, is transparently available to my to-do list app - so I can see my appointments, my tasks, and even my workouts (thanks to health kit) in one view.
Weather is a system service on iOS. Calendar is a system service. Contacts is a system service. Health data is a system service. Even music - third-party apps can control and consume your music library on iOS - is a system service (if you use it).
My Mac knows how to develop RAW files out of the box, and it can connect with integrations to get them printed, and use my library metadata to sync my favorites to my phone. I have a third party app that syncs my Apple photos library to my NAS.
A good OS, like iOS or macOS, could be the fabric that holds all your vibe-coded and ephemeral LLM apps together. I see it making a whole new category of things possible.
You still gotta write the post, though, because otherwise what you get is a circulating Twitter post.
I'll write something more fulfilling about this next week, I think.
Honestly, as a long time Emacs user, agents are amazing because when all your functionality and data lives in open source code and text files, the world is literally your oyster.
The advantage of the Apple approach is that the user’s data is instantly available across third party apps. For example, instead of needing to build (or import) a whole CalDAV apparatus to be able to build a little calendar-based todo app, you can just use the system calendar API and let the OS do the heavy lifting, making your app much more lightweight and instantly usable with almost no user configuration. This naturally applies for scripts and agent skills, too.
Because Linux doesn’t offer a rich set of system services built around the needs of desktop users and exposed through curated APIs.
Linux can’t be better at providing a shared calendar API or a shared photos library API, or one of dozens of other APIs because Linux just doesn’t have those features.
Some particular DE could offer those features, but then who is going to build the app ecosystem around that? And isn’t it going to suck when you want to make a Linux native photos app but it doesn’t work on Ubuntu because your LLM was trained on KDE?
My photos are not text files.
> partition different applications off from each other, and carefully control how they can communicate.
without
> provide access to hardware resources
That’s still definitely what an OS does. The application model is often very loosely coupled. See web browsers.
A difference between embedded systems and operating systems is the ability to provide access to hardware (in such a way as to isolate and dictate communication) _dynamically_ with a-prior unknown requests.
> Is there a meeting happening anywhere right this minute that I’m supposed to be in?
> You can go on and on like this. These all sound trivial. That’s the point. They’re normal-life problems nobody would have built serious programs around, because the audience sounds too small.
Waze and Calendars. Both have huge audiences.
It sees a small side road and thinks “I will take the user down here because there’s no traffic” But it doesn’t consider that the side road is empty because you’ll have to turn across four lanes of traffic in rush hour. So you get stuck for twenty minutes.
If your answer is you tell it, then the obvious solution to improve the app is to provide a way of specifying a road exclusion list, and that is extremely feasible.
It's an app with an audience of just me. That's the point. You're not supposed to want it. You have your own problems.
Just with the bit I quoted above, that seemed like a big misread of the market and the implications of everyone writing their own apps.
In this future, are people paying the API pricing of traffic providers? In this future, who's supplying the traffic data if no one is using their mapping apps?
I think this is coming across as glib or crabby but there's something more significant I'm trying to communicate. This subthread is in one of the classic software developer problem solving modes: pose a question or a problem, respond with "what are you really trying to do", argue that the problem was misconceived to begin with.
Eric Raymond wrote a whole giant piece about this.
But the world in which nerds could congratulate themselves about knowing what the real questions were and separating them from the dumb questions being asked by noobs who aren't holding things right is gone. Why do I care whether a software developer thinks I'm holding the tool right? Why do I care that someone familiar with Apple's fixed-function prefab software ecosystem believes there's a better way to handle my commute than what I'm trying to do?
OK? You use those tools. I'm just going to have my computer do the thing the way I want it to work. It's my commute, and my computer, this is the very specific tool I want, and all it took to get it on my screen and working reliably was for me to ask an agent, which didn't lecture me about the proper way to ask the question or what question I should have asked instead, but instead just fucking did the thing I asked for.
It's wild to me that software nerds believe they're going to win this argument.
You might as well just say "I think you're overestimating the general public's interest in using computers", because that's what it's going to mean to use a computer in a couple years, and way sooner than we think --- asking it to do arbitrary, random things, whatever we want in the moment, and then having that happen.
We don't even need to get better than Opus 5.5 to reach that moment. We just need Opus 5.5 to get significantly cheaper and faster. You can bet against that if you want, but I'm glad to be on the other side of that bet.
I don't mean to be snippy, but I think this is a really important feature of the discussion about personal apps, and it's worth being clear about it.
The ideation of features, their requirements, getting code for features written, testing the features, seeing how they fit in with the rest of the landscape, and maintaining those features. Same as before. Do those go away in your future? Even with the aid of AI agents, do you really think billions of people will want to go through that process every time they want a device to do something? Do you really think the younger generations who are being told AI will replace them will also embrace AI?
Similar to how even your hypothetical devices of the future still very much have an operating system (a kernel, libraries for software to interface with hardware, a standard suite of apps and tools, an environment for the user to kick things off)
> You might as well just say "I think you're overestimating the general public's interest in using computers"
I might, because I think the general public doesn't much like to "use the computer". It's what they need to use for work, or school, or how to listen to music or watch what they want to watch.
We're only just starting the general public's introduction to agentic AI with Siri and some Android devices. I'm betting there will eventually be some horror stories.
For example, some company did not like the fact that the call stack could be reverse engineered … yet they would not help you build a prototype at the pace of development (and this was back in 2011)
Then once you have a working OS, the second hurdle was to demonstrate a very high number of active users, such that they dare to even consider your new environment as a potential target. This a bootstrap challenge which is hard to achieve. To get traction from App developer implied getting traction from users, which implied having apps the users care about …
While you goal of removing the application layer sounds like a nice way to work-around the bootstrap problem, I fear you will still have it at the API level.
I wish you good luck, and I hope the landscape has changed in the last decade.
And to them software is generally a not much more than a home-screen bookmark to some client that could be a website.
"Write an app to provide psychological advice to people with mild anxiety, then market it and sell it to people with mild anxiety."
If an entrepreneur did that with no domain expertise, imagine the potential liability.
How do you know the output is correct for anything that matters more than completing simple and repetitive tasks? You still have to know an expert in a domain has vetted the software. The result of applying AI to OS development is making itself fairly clear with MS I believe.
It seems to me as if the current entrepreneur playbook largely is to run really fast and hope the judicial system doesn't catch up.
I find this unlikely. It's far more likely that every good idea that anyone has will just be more quickly folded into software from the big software houses.
AI doesn't slow this down, because big software houses have much more money for compute and tokens.
You have a good idea for a communication add-on? Two weeks later it's available in Teams, Slack and WhatsApp.
People aren't writing their own software, for the same reason they aren't 3D printing their own items.
Everything in the shell and you can click it. I'm building custom needle2 models to operate the panels and tabs instead of using tokens and md files
Hah, shots fired! https://steve-yegge.medium.com/why-i-left-google-to-join-gra... (Which to be fair was a prescient critique of Google too.)
What a load of self-aggrandizing bullshit. How do we keep minting these guys with a completely confabulated sense of perspective?
Nothing about this is groundbreaking and could have been written in BASIC.
I grew up on BASIC using a BBC Micro in the 80s, using a QWERTY keyboard and a screen. Some decades later I'm using a QWERTY keyboard and a screen with a few more pixels. I'm betting nothing will change much. An OS will do OS things, just a lot faster and more securely.
Spreadsheets will still be useful for making deterministic calculations about finances. It will just no longer take 100 engineers to build one.
GIS software will continue to use deterministic algorithms for calculating areas because it's a useful function but, hopefully it will get more user friendly.
I believe the author is misguided about A.I.'s capabilities just like many analysts and investors are.
A.I. (or should I say LLMs) is useful for specific problems but contrary to its name it has no "intelligence."
People could already vibe-code simple products today, but there's hardly any non-programmers that do. This is even true for businesses. For the last year people have been boasting about canceling their SaaS products, but nothing seems to have changed. SaaS stocks got hit by the ai hysteria in January, but many are back to previous levels and some even higher. There's an Indiehacker on Twitter whose product is a simple stage timer. It's not a hardware product. It's literally just the software for a stage timer. One would expect his business to have died already, but he's apparently doing completely fine.
And self-awareness for bonus points.
Most people, and I wouldn't be surprised if it was 90%, who've had any interaction with AI are still stuck on chatting with a chatbot. They've got no clue how far you can now go with a few well written markdown files and/or prompts, all in natural language!
So while I mostly agree on your outlook, and think it's super exciting, I've also felt firsthand that building with AI hasn't yet diffused as much I'd hoped it would. It's still concentrated in the hands of people who have an aptitude for building in some way, shape or form.
The question is: How do you get more people, who aren't power users or builders, to start building with AI? Maybe it starts with getting them to see that building to solve their own life's problems makes life more interesting?
- Providing a memory allocator
- Providing threads which manage how the CPU cores are used
- Interacting with network hardware, the motherboard's RTC, memory address mapping etc
- Scheduling arbitrary software to run simultaneously
- Providing a file system
- etc
This gets at the core of it; i.e. you don't need anything in that list above to run software, but it's useful when you want third party software to run on the hardware, and multiple pieces of software co-existing with each other, and running without special hardware knowledge.
So, to this point, I agree with the article/author on this distinction. I think differences arise from how different people use computers, and perhaps I quibble with the title more than the contents. There is many software I run and write which I imagine will continue relatively unchanged (or changed incrementally) for decades. Broadly, GUI creative software, media consumption etc. CAD, EDA, structural biology/bioinformatics software, games, media consumption (video, text etc), IDEs. I suppose another highlight is that I never personally understood the appeal of terminal-based software that many software programmers love. (Not just software programmers: bioinformaticians love CLI-based workflows too, i.e. piping stdin/out around)
I would love Android/iPhone competition! Although I imagine the same obstacle of "How do I get my societal-connections functionality to work in a way where I'm not excluded" will continue to be a problem in the short term, at least. (Banking, communications, NFC payments etc)
Sounds like this is a more serious and phone-centered take on what Steve demoed. We sure live in interesting times.
The hand-held computer tells them what they're permitted to do, and they conform.
Sometimes, one of the users has a non-conforming idea and says "I'd like to be a superuser on my computer" and they're shouted down with a chorus of "you'll be hacked" and "nobody will be able to trust your computer and you'll be excluded from online services". Sometimes the organisations saying this even use the excuse that "restricting the user is essential to maintaining the user's privacy".
If you look at software like sophisticated database kernels they bypass the OS almost completely because they can reimplement a better OS in userspace, using the kernel as little more than a device driver. Over time this has become more common, not less. OS like Linux have actively embraced this role.
Being able to effectively specialize the OS for specific applications has large benefits. It is a specialized skill but not unusually expensive to implement.
Uncharitably... because eventually they hope to produce a phone that isn't running iOS or Android?
More charitably because the title is actually an important question right now in the abstract for almost everything, not just operating systems.
But the forces that shape the question are so large and hard to get a handle on when we are living through the moment that you have to start thinking about it somewhere, why not OSes?
Roblox is a game where you make the game. This sounds like an OS where you make the OS.
Probably could have spelled it out more, being too pithy perhaps
Incredible how anti-intellectualism keeps taking over the software engineering profession.
If this VUI (Vibed User Interface) future comes about it entirely dismantles Apple's (and Android's) App Store moat. From a cynical perspective this may be why Apple may be shifting their strategy to enshittification: https://techcrunch.com/2026/09/06/phil-schillers-app-store-e...
Google has seen the future, and like their strategy with search (but unlike Apple, which continues blocking and suing "everything" app makers) are willing to disrupt themselves: https://techcrunch.com/2026/05/12/googles-create-my-widget-f...
However, if the slew of AI wearable or "companion" devices coming out are any indication -- from the failed Humane Pin, to Meta's keychain and glasses, to Microsoft's office badge (which OMG is so on-point in its enterpriseyness) to whatever Jonny Ive is cooking up at OpenAI -- the smartphone form factor may not be the eventual one we settle on for AI. Something to keep in mind when designing the OS.
Then the AI will just create the image that resembles a spreadsheet or whatever on the fly and write straight to the framebuffer.
All other tasks that go with the spreadsheet will also be done on the fly.
But I could imagine our language will change and certain words will basically act like macros and create a set of assumptions and specifications for the prompt that you're using.
I'm driving some place new, but of course, I first have to go through all my neighborhood streets to the freeway, and then later, eventually, something unfamiliar. I call up Google maps and it calls out all the play by play turns and such, starting from my house, instead of just saying get on the I5 North for 70 miles, then shuts up until I am close to the unfamiliar exit. It's why I mute it, I hate the blow by blow through familiar territory. Google knows where I go, God damn them, why don't they use that information already?
* improved authorization, transparency and control. Presently OS's have privileged and unprivileged modes . AI has too many privileges in unprivileged modes (all credentials, all files). better control over delegating authority is needed.
* improved interface consistency. UIs still inform people on the capabilities and intentions of the machine. A blank text box with an agent is not informative. How many people have the agency to turn the blank box into a creation. A UI illustrates the toolbox available to people and the machines broader capabilities
(The irony is that all the reading that's gone into our design is like, old CSAIL papers. Maybe it's true! Maybe everything we're doing is just crate-digging through old settled computer science. That would be on brand. I should find a way to work Athena into it. Maybe we'll use Zephyr for messaging.)
I guess it really is over if this is what professionals (supposedly) seem to think.
OP is right though, an OS remains an OS, same as it’s always been. What’s changed is how we interact with it and what we lay on top of it. Nothing magical or philosophical about it.
The most recent was Loris Degioanni yesterday in a conference talk.
I think saying " This dumb dumb doesn't know what a device driver is" is a very incorrect read since you know, he's written a bunch of them
Finding some balance between access and safety for vibe-conjuring parts of the OS in a way that's still powerful but won't accidentally nuke or corrupt itself.
It's the thing that runs your AI agent, that runs your AI pocket bot (AIBI looks cute btw), that runs the software doing the training of AI models, that runs the datacenters doing the inference, etc.
There's still data. There are still bits moving. Filesystems and networking aren't going anywhere.
In a charitable reading the title is something like "what is even a desktop OS now?" or "what is even a consumer OS now?".
As for the new project, I've got one bet: there's Linux involved to host it.
Linux is an OS.
P.S: as a sidenote the entire "users don't install apps anymore" was already true to some extent. A shitload of users are totally fine with an OS that runs a browser and that's it. And they're fully operational. And we still all know what an OS is, even for, say, a Chromebook. Heck, I'm not even sure "installing apps" as ever been, except during the first install of the OS, what an OS was ever about.
https://www.proactiveinvestors.com/companies/news/1082001/te...
So, what are you really working on?
Really reads like an inner dialogue or when someone wants to sell me something.
Most of these problems the slop apps can solve are dependent on gathering information to answer questions. If no one ever visits weather.com to get weather information, ad revenue will completely dry up and the site will go under. Then the slop stops slopping and I can’t get my hyper local weather app to work.
I’m not going to go as far as to say this article is wrong, in fact in this cursed timeline it’s probably right. I just don’t see how it’s good.
On the other hand, that’s just traditional banner/video style advertising where the goal is to force specific content on the user. If someone wants to spread the word about their product or service, I’m sure they’ll pay money to advertising services that use agents to inject their message into the global context. This is happening already, we just talk about it in the frame of “dead internet theory” and AI influencers, etc. Anywhere that content is being produced and consumed is an opportunity for the message to be nudged in favor of the person paying money.
Like many major cultural phenomenon, expect a pendulum effect, more than a beginning or end of a trend.
Hubris detected.
In all seriousness though, I'm not so sure.
Power users are lazy users.
And we're not all serial entrepreneurs with 400k karma points on HN, remember.
But really, a platform that allows one to do all levels off apps in this way where one can cleanly reset their side effects and effect on the system, and the app is exposed just as an icon, not a folder of source files, could already be possible and could lower the bar of starting vibe coding to the bottom
Back when Google was the newest thing, I would Google all the obscure things that I could think of, trying to fill in gaps in my knowledge. Trying to verify urban myths but usually debunking them, etc.
Any power user or even any user, could have done that. But most people who considered themselves critically thinkers actually did the opposite, digging deeper into all the harmful rabbit holes we see today.
A lot of things can done "with just a few steps" with AI but I think it's underestimated how hard it is to come up with those steps.
Not that the article isn't interesting otherwise.
IMO even this AI phone is doomed in the slightly longer run.
One more step into the weirdness and people will just be working with AI to build their own bespoke phone with the specs sent over to a manufacturer that is "like PCBWay, but spits out whole devices".
I have no idea what this does to society, the economy, etc when we reach a point where nobody really "needs" any kind of middleman for almost anything (well, except for things they actually really need, like food and shelter), but it looks like I'll see it happen for a bit before I die.
The physical device that is best for me is probably the best car compromise for several million other people.
if my ex calls and I want my glass brick to flash the "help me" in SOS on the screen, play Hector Berlioz, and dim all the lights in the room – I can have that. I don't have to want the same thing a million other people do.
First of all the initial definition is totally correct, and if your embedded function needs to call an intermediate layer to access the hardware that layer is an OS. Second of all, all those "purposes" are exactly achieved by providing _managed_ access to hardware resources.
And that is valuable in whatever world we end up living in. Sure, 10 years from now my bank app and my Hacker News browser may both be written by "my LLM" and tailored to my use cases and desires. Will I want them to share memory? I don't think so.
Software in the future will have to be building blocks for agents to extend. For my projects I decided to generate LLM targeted guides in the docs/ folder, to teach agents quickly "how to" use the project as a framework.
How to implement an MVC view, where to place the files, a guide how to write a good backend route, how to extend the schemas, and other articles like that.
In my opinion we've already reached "Star Trek like computers". Our OS will have to accommodate for better sandboxing and better permission management and better role separation policies, because currently that whole security sector assumes that programs will be static and won't change before/after an approval.
We need to figure out how to better sandbox programs in terms of not only control flow, but also in terms of access to _data_ on a user's behalf.
This seems to be only thought out after a day of vibe coding apps that no-one uses.
Unclear what will be left next to it. It will just live on your phone and take care of everything. The apps will be markdown files describing what they do, probably compiled ahead of time for latency.
But the model will be in control, so what else is left to implement besides giving it bash, a compiler and a display?
I guess there is value in the year before that happens.
The same. The slop apps remain just that. They may do something useful for you, but that's it. A bit like a branch in a repository.
Yes, I'm sure we all have horror stories about spreadsheets that got out of hand, but that ignores the thousands (or more) spreadsheets that are silently productive everyday.
not just startup nerd, he also make gas town and drunken rant