Strange, it works here in Firefox on Linux using the internal GPU (I don't use primusrun for the browser). Normally I'm the first to notice a particularly heavy website or bad FPS in even pretty old games! Wonder how they managed to make it sluggish on a very expensive GPU but get my crappy setup to run it nicely
Odd. My laptop seemed to do fine with a 'NVIDIA GeForce GTX 1650 Ti Mobile [Discrete]' using CachyOS. It could have been a little smoother but it rendered fine. There were a couple spots where it was a little herky-jerky-laggy that maybe needs optimization.
It choked on my M5 Max MBP w/ 128GB of Unified Memory connected via a TB5 dock over 10GbE directly to my router, which is backed by 5Gbit symmetric fiber. My measured average latency to nasaforce.gov was under 13ms.
There seems a fair enthusiasm in the UI of these to hide code from coders. Like the prompt interaction is the true source and the actual code is some sort of annoying intermediate runtime inconvenience to cover up. I get that productivity can be improved with a lot of this for non developers, just not sure using 'code' as the term is the right one or not.
> There seems a fair enthusiasm in the UI of these to hide code from coders. Like the prompt interaction is the true source and the actual code is some sort of annoying intermediate runtime inconvenience to cover up.
I've finally started getting into AI with a coding harness but I've take the opposite approach. usually I have the structure of my code in my mind already and talk to the prompt like I'm pairing with it. while its generating the code, I'm telling it the structure of the code and individual functions. its sped me up quite a lot while I still operate at the level of the code itself. the final output ends up looking like code I'd write minus syntax errors.
This is the way to do it if you're a serious developer, you use the AI coding agent as a tool, guiding it with your experience. Telling a coding agent "build me an app" is great, but you get garbage. Telling an agent "I've stubbed out the data model and flow in the provided files, fill in the TODOs for me" allows you the control over structure that AI lacks. The code in the functions can usually be tweaked yourself to suit your style. They're also helpful for processing 20 different specs, docs, and RFCs together to help you design certain code flows, but you still have to understand how things work to get something decent.
Note that I program in Go, so there is only really 1 way to do anything, and it's super explicit how to do things, so AI is a true help there. If I were using Python, I might have a different opinion, since there are 27 ways to do anything. The AI is good at Go, but I haven't explored outside of that ecosystem yet with coding assistance.
If you use a type checker in strict mode (e.g. pyright with "typeCheckingMode: strict") and a linter with strict rules (e.g. ruff with many rules enabled), the output space is constrained enough that you can get pretty consistent Python code. I'm not saying this is "good Python" overall, but it works pretty well with agents.
My workflow is quite similar. I try to write my prompts and supporting documentation in a way that it feels like the LLM is just writing what is in my mind.
When im in implementation sessions i try to not let the llm do any decision making at all, just faster writing. This is way better than manually typing and my crippling RSI has been slowly getting better with the use of voice tools and so on.
The funny thing is my expectation was that adoption of AI coding would kill the joy of getting into a flow state but I've actually found myself starting to slip into an alternate type of flow state.
Instead of hammering out code manually over an hour the new flow state is a back and forth with the LLM on something that's clear in my mind. It's a collaborative state where I'm ultimately not writing much code manually but I'm still bouncing between technical thoughts, designing architecture, reviewing code, switching direction etc.
Yeah - similar thing for me as well. A lot of times there would be something I want to work on that would be boilerplate/repetitive/laborious work and I would just procrastinate it for as long as possible, working on other things, until I'd finally get around to doing it. Now those are just immediately completed with a simple prompt and instead of going with the initial implementation, I have the bandwidth to tweak and refine details that I would have skipped over before just to ship.
I personally have been finding good results "hiding the code" behind the harnesses. I do have to rely on verification and testing a lot, which I also get the AI to do, but for most of the cases it works out well enough. A good verification and testing setup with automated, strict reviewing goes a long way.
Yes and the official docs even mention that if you’re on Windows you should run Codex CLI via WSL. Meaning, it’s specifically designed for unix systems.
The power to the people is not us the developers and coders.
We know how to do a lot of things, how to automate etc.
A billion people do not know this and probably benefit initially a lot more.
When i did some powerpoint presentation, i browsed around and draged images from the browser to the desktop, than i draged them into powerpoint. My collegue looked at me and was bewildered how fast I did all of that.
I've helped an otherwise very successful and capable guy (architect) set up a shortcut on his desktop to shut down his machine. Navigating to the power down option in the menu was too much of a technical hurdle. The gap in needs between the average HNer and the rest of the world is staggering
This. I’m sure everyone has a similar story of how difficult it was to explain the difference between a program shortcut represented as a visual icon on a desktop versus the actual executable itself to somebody who didn’t grow up in the age of computing. And this was Windows… the purported OS for the masses not the classes.
Initially I thought you meant “software architect” and I was flabbergasted at how that’s possible. Took me a minute to realize there’s other architects out there lol.
Oh boy, the gap between the average it professional and ai pros here is already staggering, let alone the rest of the world. I feel like an alien, no matter where.
Yes! Even closing the windows of programs that users no longer need is hard.
It's easy to develop a disconnect with the level that average users operate at when understanding computers deeply is part of the job. I've definitely developed it myself to some extent, but I have occasional moments where my perspective is getting grounded again.
I don't think that's representative of most non-CS professionals. Most people in the fields I know (mostly professors, medical doctors, and businesspeople) can use google chrome, word, powerpoint, and a little of excel decently. There are the occasional few who confuse spreadsheets and databases, but no one who thinks shutting down computers or closing windows is hard. Heck, my ageing dad managed to troubleshoot his printer without any help, and he has no formal computer experience whatsoever.
HN has a long history of patronising the "average user" in the guise of paternal figures who don't realise that what they are doing is belittling the vast majority of tech users. I'm guilty of it myself. But they're capable of a lot more than we think they are.
Ultimately, it comes down to the willingness people have to learn new things. If they're curious enough to think about how things work, they'll be fine.
I'm not doing this to be patronising, more like telling people or myself that assumptions i make, are just not necessarily true for everyone.
And weirdly enough, a Task like sorting a file with data in it, if you are not a professional, windows offers very if not non single way of doing this. You would need to understand file types, understand that csv can be imported for excel, you need excel, than you need to understand excel how to sort stuff in it.
The ffirst thing I do in Excel is select the pseudo table and click on table -> insert to make it a sortable / real table. I showed this to every one in my Team full of studied CS people because non of them knew this.
Well, I didn't mean for this to be patronizing, but rather as a warning that not everybody is at the same level and the spread is huge. I see it often enough.
It's a while since I've used Windows but I seem to remember it giving a choice of sleep, logout, switch session etc. I could totally see someone wanting a single button for it.
Yes, the code is still important. For example, I had tasked Codex to implement function calling in a programming language, and it decided the way to do this was to spin up a brand new sub interpreter on each function call, load a standard library into it, execute the code, destroy the interpreter, and then continue -- despite an already partial and much more efficient solution was already there but in comments. The AI solution "worked", passed all the tests the AI wrote for it, but it was still very very wrong. I had to look at the code to understand it did this. To get it right, you have to either I guess indicate how to implement it, which requires a degree of expertise beyond prompting.
Yep, all models today still need prompting that requires some expertise. Same with context management, it also needs both domain expertise as well as knowing generally how these models work.
Do you ask it for a design first? Depending on complexity I ask for a short design doc or a function signature + approach before any code, and only greenlight once it looks sane.
I understand the "just prompt better" perspective, but this is the kind of thing my undergraduate students wouldn't do, why is the PhD expert-level coder that's supposed to replace all developers doing it? Having to explicitly tell it not to do certain boneheaded things, leave me wondering: what else is it going to do that's boneheaded which I haven't explicit about?
Because it's not "PhD-expert level" at all, lol. Even the biggest models (Mythos, GPT-Pro, Gemini DeepThink) are nowhere near the level of effort that would be expected in a PhD dissertation, even in their absolute best domains. Telling it to work out a plan first is exactly how you would supervise an eager but not-too-smart junior coder. That's what AI is like, even at its very best.
That's not the best framing, IMO. More important is, even a PhD expert human wouldn't one-shot complex programs out of short, vague requests. There's a process to this. Even a thesis isn't written in one, long, amphetamine-fueled evening. It's a process whose every steps involves thinking, referencing sources, talking with oneself and other people, exploring possibilities, going two steps forward and one step back, and making decisions at every point.
Those decisions are, by large, what humans still need to do. If the problem is complex, and you desperately avoid needing to decide, then what AI produces will surprise you, but in a bad way.
I understand that but 1) expert-level performance is how they are being sold; but moreover 2) the level of hand-holding is kind of ridiculous. I'll give another example, Codex decided to write two identical functions linearize_token_output and token_output_linearize. Prompting it not to do things like that feels like plugging holes in a dyke. And through prompting, can you even guarantee it won't write duplicate code?
I'll give a third example: I gave Codex some tests and told it to implement the code that would make the tests pass. Codex wrote the tests into the testing file, but then marked them as "shouldn't test", and confirmed all tests pass. Going back I told it something to the effect "you didn't implement the code that would make the tests work, implement it". But after several rounds of this, seemingly no amount of prompting would cause it to actually write code -- instead each time it came back that it had fixed everything and all tests pass, despite only modifying the tests file.
In each example, I keep coming back to the perspective that the code is not abstracted, it's an important artifact and it needs/deserves inspection.
Note to self: don't go away for the weekend without HN ;-)
Personally I run several agents. At minimum Codex and Claude, so they cross-check each other. Exactly to avoid what you describe. Duplicate functions or "tests all pass" when nothing was implemented is the kind of thing a second model, reading the diff with fresh eyes, tends to flag immediately.
But it takes coordination and solid skills + rules to make it work. Tomorrow's battle -> AI skillers.
> the code is not abstracted, it's an important artifact and it needs inspection.
That's a rather trivial consideration though. The real cost of code is not really writing it out to begin with, it's overwhelmingly the long-term maintenance. You should strive to use AI as a tool to make your code as easy as possible to understand and maintain, not to just write mountains of terrible slop-quality code.
I think this would work much better if there were constraints in place, a software stack clearly separating different concerns - e.g. you just ask AI to write business logic while you already have data sources, auth, etc, configured.
But that's not how popular, modern software stacks work. They are like "you can do anything, anything at all!".
Consider Visual Basic for Applications - normally your code is together with data in one document, which you can send to colleague. It can be easily shared, there's nothing to set up, etc.
That's not true for JS, Python, Java, etc - you need to install libraries, you need to explicitly provide data, etc. Software industry as a whole embraced complexity because devs are paid to deal with complexity.
Now AI has to use same software stacks as the rest of the industry, making software fragile, requiring continuous maintenance, etc. VBA code which doesn't use any arcane features would require no maintenance and can work for decades.
So my guess is that the bottleneck might be neither models nor harness/wrapper - but overall software flimsiness and poor architectural decisions
In UI, I’m pretty sure that replacement is already here. We’ll be lucky if at least backend stays a place where people still care about the actual source.
I'd say the opposite, the frontend code is so complex these days that you can't escape the source code.
If you stick to tailwind + server side rendered pages you can probably go pretty far with just AI and no code knowledge but once you introduce modern TS tooling, I don't think it's enough anymore.
I knew a guy who did 6510 and 68000 assembler for many years and had a hard time using higher order languages as well as DSLs. “Only assembler is real code. Everything else is phony, bloat for what can be done way better with a fraction of the C++ memory footprint.”
Well that guy was me and while I still consider HOLs as weird abstractions, they are immensely useful and necessary as well as the best option for the time being.
SQL is the classic example for so called declarative languages. To this day I am puzzled that people consider SQL declarative - for me it is exactly the opposite.
And the rise of LLMs proof my point.
So the moral of the story is, that programming is always about abstractions and that there have been people, who refused to adopt some languages due to a different reference.
The irony is, that I will also miss C like HOLs but Prompt Engineering is not English language but an artificial system that uses English words.
Abstractions build on top of abstractions. For you code is HOL, I still see a compiler that gives you machine code.
As a child I couldn't understand why I have to talk in a cryptic language and can't just write a for loop when working with DBs. In hindsight it was a valuable lesson that implementation details matter even though I wouldn't want them to.
I think the intent is more "we won't need coders" ... the real goal is to get to the point where Product Managers can just write specs and a working product comes out the other end.
These people HATE that developers have been necessary and highly paid and, in their view, prima donnas. I think most of the people running these companies actually despise developers.
Code isn't going anywhere. Code is multiple orders of magnitude cheaper and faster than an LLM for the same task, and that gap is likely to widen rather than contract because the bigger the AI gets the sillier it gets to use it to do something code could have done.
Compare the actual operations done for code to add 10 8-digit numbers to an LLM on the same task. Heck, I'll even say, forget the possibility the LLM may be wrong. Just compare the computational resources deployed. How many FLOPS for the code-based addition? How many for the LLM? That's a worst-case scenario in some ways but it also gives you a good sense of what is going on.
Humans may stop looking at it but it's not going anywhere.
I think grandparent comments were talking about how Codex designers try to push LLMs to displace the interface to code, not necessarily code itself. In that view, code could stay as the execution substrate, but the default human interaction layer moves upward, the way higher-level languages displaced direct interaction with lower-level ones. From a HCI perspective, raw computational efficiency is not the main question; the bottleneck is often the human, so the interface only has to be fast and reliable enough at human timescales.
Everyday people can now do much more than they could, because they can build programs.
The idea that code is something sacred and only devs can somehow do it is dying, and I personally love it, as I am watching it enable so many of my friends and family who have no idea how to code.
Today, when we think of someone "using the computer" we gravitate towards people using apps, installing them, writing documents, playing games. But very rarely have we thought of it as "coding" or "making the computer do new things" -- that's been reserved, again, for coders.
Yet, I think that a future is fast approaching where using the computer will also include simply coding by having an agent code something for you. While there will certainly still be apps/programs that everyone uses, everyone will also have their own set of custom-built programs, often even without knowing it, because agents will build them, almost unprompted.
To use a computer will include _building_ programs on the computer, without ever knowing how to code or even knowing that the code is there.
There will of course still be room for coders, those who understand what's happening below. And of course that software engineers should know how to code (less and less as time goes on, though, probably), but no doubt to me that human-computer interaction will now include this level of sophistication.
> I am watching it enable so many of my friends and family who have no idea how to code.
Be careful what you wish for, this is going to be a double edged sword like YouTube is. YouTube allowed regular people without money and industry connections to make all sorts of quality, niche content. But for every bit of great content, there’s 1000 times as much garbage and outright misleading shit.
Giving people without any clue how computing works the ability to create software that interfaces with the outside world is likewise going to create some great stuff and 1000 times as much buggy and dangerous stuff. And allow untold numbers of scammers with no technical skill the ability to scam the wider world.
I'm aware, and I'll very much take those odds. This is just another problem for humanity to solve in its quest to empower itself.
I'm not sure how we're going to solve the obviously relevant problem of slop, but I would rather die trying, than restrict access to knowledge and capability because of evil. I believe in the GOOD of humanity. We WILL find a way.
> The idea that code is something sacred and only devs can somehow do it is dying, and I personally love it, as I am watching it enable so many of my friends and family who have no idea how to code.
People on HN are seriously delusional.
AI removed the need to know the syntax. Your grandma does not know JS but can one shot a React app. Great!
Software engineering is not and has never been about the syntax or one shotting apps. Software engineering is about managing complexity at a level that a layman could not. Your ideal word requires an AI that's capable of reasoning at 100k-1 million lines of code and not make ANY mistakes. All edge cases covered or clarified. If (when) that truly happens, software engineering will not be the first profession to go.
I never said Software Engineering is dying or needs to go. I'm not the least bit afraid of it.
In fact, in the very message you're replying to, I hinted at the opposite (and have since in another post stated explicitly that I very much think the profession will still need to exist).
My ideal world already exists, and will keep getting better: many friends of mine already have custom-built programs that fit their use case, and they don't need anything else. This also didn't "eat" any market of a software house -- this is "DIY" software, not production-grade. That's why I explicitly stated this is a new way of human-computer-interaction, which it definitely is (and IMO those who don't see this are the ones clearly deluded).
> Everyday people can now do much more than they could, because they can build programs.
Indeed. Just spoke to a buddy, he's got some electronics knowledge, he's been code-curious but never gotten past very simple bash scripts and Excel sheets (vlookup etc to drive calculations).
He got himself a Claude subscription and has now implemented a non-trivial Arduino project, involving multiple CAN-bus modules and an interactive, dynamic web interface to control all this. The web interface detects the CAN-bus modules and populates the web interface based on that, and allows him to adjust the control logic.
It's a project he's had in his head for a few years and now was able to realize on his own (modulo Claude).
Exactly the kind of thing I've been seeing too. And often with people who know even less.
You spoke of an Arduino, and I have a friend with zero coding knowledge who built a fun project with an ESP32 and a tiny camera to detect when they are "not looking at the computer".
But, sure, people keep saying we're delusional when we say that this is where the world is headed: people building things, so often without even knowing they are doing anything "different" than what they were doing before, when they simply clicked buttons and "things just happened in the computer".
Not a musician so perhaps not accurate but I feel it's a bit like synths and DAWs. You might not be good at playing an instrument, something which requires a certain dexterity, finesse and lots of training. But with some virtual synths and a DAW you can make music that people enjoy.
i WISH we weren't phoning with them anymore, but people keep trying to send me actual honest-to-god SMS in the year 2026, and collecting my phone number for everything including the hospital and expect me to not have non-contact calls blocked by default even though there are 7 spam calls a day
In what world would I prefer to give someone access to me via a messaging app rather than a fully-async text SMS message? I don't even love that people can see if you've read their texts now.
I believe that in all of South America people exclusively use WhatsApp to communicate via text because SMS is only used for spam and bad 2FA. Companies are even using WhatsApp for 2FA now instead of SMS and the fact that americans use SMS is viewed as a joke.
It's not the code I write, it's what I've noticed from people in 25 years of writing code in the corner.
All of my friends who would die before they use AI 2 years ago now call themselves AI/agentic engineers because the money is there. Many of them don't understand a thing about AI or agents, but CC/Codex/Cursor can cover up for a lot.
Consequently, if Claude Code/"coding agents" is a hot topic (which it is), people who know nothing about any of this will start raising money and writing articles about it, even (especially) if it has nothing to do with code, because these people know nothing about code, so they won't realize what they're saying makes no sense. And it doesn't matter, because money.
Next thing you know your grandma will be "writing code" because that's what the marketing copy says. That's all it takes for the zeitgeist to shift for the term "code". It will soon mean something new to people who had no idea what code was before, and infuriating to people who do know (but aren't trying to sell you something).
I know that's long-winded but hopefully you get where I'm coming from :D.
Totally this. People who don't see this seem to think we're in some sort of "bubble" or that we don't "ship proper code" or whatever else they believe in, but this change is happening. Maybe it'll be slower than I feel, but it will definitely happen. Of course I'm in a personal bubble, but I've got very clear signs that this trend is also happening outside of it.
Here's an example from just yesterday. An acquaintance of mine who has no idea how to code (literally no idea) spent about 3 weeks working hard with AI (I've been told they used a tool called emergent, though I've never heard of it and therefore don't personally vouch for it over alternatives) to build an app to help them manage their business. They created a custom-built system that has immensely streamlined their business (they run a company to help repair tires!) by automating a bunch of tasks, such as:
- Ticket creation
- Ticket reporting
- Push notifications on ticket changes (using a PWA)
- Automated pre-screening of issues from photographs using an LLM for baseline input
- Semi-automated budgeting (they get the first "draft" from the AI and it's been working)
- Deep analytics
I didn't personally see this system, so I'm for sure missing a lot of detail. Who saw it was a friend I trust and who called me to relay how amazed they were with it. They saw that it was clearly working as intended. The acquaintance was thinking of turning this into a business on its own and my friend advised them that they likely won't be able to do so, because this is very custom-built software, really tailored to their use case. But for that use case, it's really helped them.
In total: ~3 weeks + around 800€ spent to build this tool. Zero coding experience.
I don't actually know how much the "gains" are, but I don't doubt they will definitely be worth it. And I'm seeing this trend more and more everywhere I look. People are already starting to use their computer by coding without knowing, it's so obvious this is the direction we're going.
This is all compatible with the idea of software engineering existing as a way of building "software with better engineering principles and quality guarantees", as well as still knowing how to code (though I believe this will be less and less relevant).
My experience using LLMs in contexts where I care about the quality of the code, as well as personal projects where I barely look at the code (i.e. "vibe coding") is also very clearly showing me that the direction for new software is slowly but surely becoming this one where we don't care so much about the actual code, as long as the requirements are clear, there's a plethora of tests, and LLMs are around to work with it efficiently (i.e. if the following holds -- big if: "as the codebase grows, developing a feature with an LLM is still faster than building it by hand") . It is scary in many ways, but agents will definitely become the medium through which we build software, and, my hot-take here (as others have said too) is that, eventually, the actual code will matter very little -- as long as it works, is workable, and meets requirements.
For legacy software, I'm sure it's a different story, but time ticks forward, permanently, all the time. We'll see.
From what you describe, I probably would have charged them a tad more and taken a tad longer to deliver. However they would receive a production-ready application, that properly filters and sanitises and normalizes input, that is robust and resilient and reasonably extensible, and has a logical database format.
Tell me, does this vibe coded app running this business properly handle monetary addition, such as in invoicing or summarizing or deciding how big a check to write to the tax man? Are you sure? No floating point math hiding intermittent bugs?
That's actually a great point. The real problem we have is putting businesses and clients together. And traditional advertising is certainly not the answer.
My point was ~~two~~(edit: three)-fold (which, I guess, reading again is just the same thing said three times slightly differently...sorry!), more along the lines of:
- I don't think they need the extra you would offer them. I'm pretty sure they didn't add anything related to accounting. I also have to admit I'm a bit shocked that you would do all of what I described for "a tad more" than 900€, especially taking "a tad" longer than 3 weeks. To me, that's barely anything. But I guess I'll take your word for it.
- For many things, people no longer need the specialized production-ready work, precisely because they have this powerhouse at the fingertips. They "didn't find you" because it would make little sense to do so. It would take longer (which in some sense is higher risk), be more expensive, inherently be more likely to take even longer to really reach the right requirements (getting the knowledge out of their head and into yours would certainly add some overhead) and, in the end, it will likely really not bring in enough superiority for their use case.
- Because people don't need specialized production work, they won't even think of looking for it -- they already have the tools "at home". Why would I go out to buy a an electric screwdriver if I have a manual screwdriver at home? It's good enough. Sure, some people will try to use the manual one even when they shouldn't, but that's life: some people are better than others at figuring this shit out. I'm (slightly) hoping the AIs themselves will help people realize when they're trying to do something they shouldn't.
I truly believe that, for the most part, software engineering is not under threat. That there are many places where software engineering will continue to be essential. We're not developers and never have been. I think coding "manually" will die out, but not the knowledge of code (at least not for quite some time).
At the same time that I believe this, I also really believe that there is a sort of "new DIY" market (or a new "way of interacting with the machine") where ordinary people will just code things without needing to know how to code. Most of these won't be products, but they will be sufficient, for a sufficiently long time, for their needs. If/when they need more, they'll likely need the help of a software engineer, and that's more than fine.
I'm not saying this is the case with you (it doesn't seem like it is), but I see so much pushback from people who seem....either scared or in denial(?) about this (to me) very obvious new emerging way of interacting with a computer. People ask the computer to do things, and the computer builds programs and integrations between programs that....do the thing! When I was a kid, this would have been amazing, and I'm so excited that it exists now. And of course some of these "ordinary" people will also have this be their gateway into proper software engineering.
When I say friends and family, I mean it: they're all slowly starting to build tiny apps without knowing a single line of code. They often don't look good and have idiosyncrasies, but they're great for them. A friend of mine has a personal assistant with voice + telegram bot that edits their calendar and their notion, all deployed with railway (when they showed this to me I was gobsmacked!). They have ZERO coding experience...and yet...they have built this! I wouldn't use it (too finicky for me), but they swear by it and love it. (I audited the code after they asked me to and didn't find any security issues.)
Just like my dad used to grab a bit of scotch-tape to patch things up around the house, or like my grandpa used to build his toys, and furniture, he can now grab an AI and patch things up in his digital life and workplace -- how can people not see that this is happening? And, worse, why are they so very clearly upset about it and wishing that it just doesn't succeed? Is it job safety? The feeling that their favorite part of the job is being profoundly shaken up (coding)? I guess I can sort of understand and sympathize with feeling scared, but....not with the denial of it.
You know how so many people run their businesses off of excel spreadsheets? Often for way longer than they should, no doubt -- but they do. This is sort of the next step after that for some businesses. But, most of all, I really mean that for people's personal needs, interacting with the computer will involve the computer building some code for them to achieve their goals. Yes, MS is fumbling copilot, but one such integrated AI will eventually succeed, and people will open up their "start menu" / "copilot" / "Claude Cowork" / "whatever" and say "I want to create a library for my comic book collection", and over a couple of prompts (perhaps over a couple of days), their computers will just...build it. They will sometimes use existing solutions, but often they'll just build a good-enough thing that will be almost exactly what this person wants. And that's....awesome. So awesome that we're at a point where computers will enable people to do so much more.
I agree with just about everything you've mentioned.
> getting the knowledge out of their head and into yours
That's creating the spec, which is a significant portion of the work and the time (and thus the budget). Maybe I should suggest to potential clients to bang out a preliminary spec with their favourite AI chatbox before meeting. That could save significant time for both of us, and that's money. And it would force me to articulate exactly what value I add rather than having them press the "Code It For Me" button.
> People ask the computer to do things, and the computer builds programs and integrations between programs that....do the thing!
The computer builds a program that ostensibly does the thing. Under ideal conditions, while under negligible load, with expected inputs and a well-meaning operator. Real world software must consider malformed or malicious input, cyclomatic complexity, resource usage, atomicity, sudden loss of power, the ability to actually restore a backup, floating point math, race conditions, I unnormalized text, security, reproducibility, debuggability, logging, and so many other things.
My career is pivoting from writing software to cleaning up other people's vibe-coded software.
I actually love the vibe-coding movement as it makes custom software available to more people, and also extends my own career as I pivot to clean up the messes.
Fully agree. Non-dev solutions are multiplying, but devs also need to get much more productive. I recently asked myself "how many prompts to rebuild Doom on Electron?" Working result on the third one. But, still buggy though.
The devs who'll stand out are the ones debugging everyone else's vibe-coded output ;-)
That’s a pretty far cry from “nobody makes phone calls”. You can also find people who spend 6+ hours on phone calls everyday, including people under 30.
On the flip side, I cause a medium panic in my daughter when I text "please call me when you can" without a why attached. She assumes someone's in the hospital or dying or something.
My mom had to lay down a rule that if I called her at a weird hour I needed to open with whether or not I was okay. Almost 30 now and still do the same thing.
Agree with that, they seem really good limits for daily use on something like Chat GPT Pro $20 account. I'm in the curious situation of using the Codex CLI within Cursor IDE and not really getting value out of my $60 Cursor sub. Plus at every update it seems Cursor seems to break more of their UI in the 'not a cloud agent chat UI' vs the more traditional VSCode sort of layout of code first. I should probably cancel.
This looks interesting and I use Codex a fair bit already in vscode etc, but I'm having trouble leaving a 'code editor with AI' to an environment that sort of looks like it puts the code as a hidden secondary artefact. I guess the key thing is the multi agent spinning plates part.
I find that the case too. For more complex things my future ask would be something that perhaps formalized verification/testing into the AI dev cycle? My confidence in not needing to see code is directly proportional in my level of comfort in test coverage (even if quite high level UI/integration mechanisms rather than 1 != 0 unit stuff)
I did too - did the SF back to NYC leg that goes the north route as well. Amazing experience. I was only 19 at the time. My favorite memory is that on the westbound leg we met up with the eastbound one in (I think) Yellowstone and while parking the buses they managed to slowly crash in to each other (just a dent, nothing serious). I liked the fact they both started from separate coasts and ending up colliding.
Can I ask what year that was in? Several people in here seem to be intimately familiar with this company, but it's my first time hearing about it, despite me being quite interested in travel and all sorts of weird transit methods. It made me wonder if everyone here used them when they were at their peak many decades ago, or if this varied.
1989. I saw a paper flyer in the New York Public Library and turned up a few days later at the bus station on impulse. It was basically a moving commune, where you slept on flat boards, cooked together and moved glacially slowly towards the destination. The NY -> SF went south via FL, TX etc. It was mainly young people not actually from the US, Europe, Australia etc.
So is maybe gpt-5.2 with reasoning set to 'none' identical to gpt-5.2-chat-latest in capabilities but perhaps with a different system (system) prompt? I notice chat-latest doesn't accept temperature or reasoning (which makes sense) parameters, so something is certainly different underneath?
> ChatGPT is widely used for practical guidance, information seeking, and writing, which together make up nearly 80% of usage. Non-work queries now dominate (70%). Writing is the main work task, mostly editing user text. Users are younger, increasingly female, global, and adoption is growing fastest in lower-income countries
That's funny, the way I interpreted this sentence is that usage was already high in older, male, and high-income countries so most of the new users are coming from outside these demographics. Which, ironically, is the exact opposite of what you're saying.
You read "Users are younger, increasingly female, global, and adoption is growing fastest in lower-income countries" and gathered that "Young moms with no money in poor countries use this product the most". Do I really need to spell out the fact that you completely failed to understand basic English here?
If they mostly ask how to raise their children and follow the received advice... Then yeah, in some 20 years we'll see what kind of return we get. People raised on social media are one thing; people raised by (with the assistance of) ChatGPT may be even worse off because of it.
My initial assumption would be that there are a lot, likely a majority, of parents who have had next to no advice on how to raise kids. Furthermore, I would posit that many of them were not raised in particularly nurturing circumstances themselves.
As such, I would expect that the advice ChatGPT gives (i.e. an average from parenting advice blogs and forums), would on average result in better parenting.
That's obviously not to say that ChatGPT gives great advice, but that the bar is very low already.
You're right, as much as I'd like not to be aware of it. Indeed, the bar is very low.
Whether heeding ChatGPT advice would be better or worse than no advice at all, I honestly cannot say. On the one hand, getting some advice would probably help in many, many cases - there's a lot of low-hanging fruit here; on the other, low-quality advice has the potential to ruin the lives of multiple people at any moment. This is like medical or lawyer advice: very high stakes in many cases. Should we rely on a model that doesn't really understand the underlying logic for advice on such matters? The "average" of parenting blogs can be a mish-mash of different philosophies or approaches glued together, making up something that sounds plausible but leads to catastrophic results years or decades later.
I don't know. Parenting is a complex problem in itself; then you have people generally not looking for advice or being unable to recognize good advice. It doesn't look like adding a hallucinating AI model to the mix would help much, but I may be wrong on this. I guess we'll find out the hard way: through people trying (or not) it out and then living with consequences (if any).
No amount of LinkedEn speech can fix the poor part of it.
In 2025, it's abundantly clear that the mask is off. Only the whales matter in video games. Only the top donors matter in donation funding. Modern laptops with GPUs are all $2k+ dollars machines. Luxury condos are everywhere. McDonalds revenues and profits are up despite pricing out a lot of low income people.
The poor have less of the nothing they already have. You can make a hundred affordable cars or get as much, if not order of magnitudes more, profit with just one luxury vehicle sale.
Most political donors are $25/month Actblue donations, and it doesn't matter because the campaigns with the most donations regularly lose.
> McDonalds revenues and profits are up despite pricing out a lot of low income people.
They didn't really raise prices, they just put coupons in the app.
> Luxury condos are everywhere.
Houses don't cost more because they have "luxury" features. A nicer countertop doesn't hypnotize people into paying more for a house. Prices are negotiated between buyer and seller and most of the development cost is the land price.
> The poor have less of the nothing they already have.
Wage inequality in the US is lower than it was in 2019. In general income inequality hasn't increased since 2014.
You have no idea if they’re ambitious or educated. Absolutely no idea. Is it just commonplace to inject “facts” into conjecture? Comes off as desperate.
I get a lot of productivity out of LLMs so far, which for me is a simple good sign. I can get a lot done in a shorter time and it's not just using them as autocomplete. There is this nagging doubt that there's some debt to pay one day when it has too loose a leash, but LLMs aren't alone in that problem.
One thing I've done with some success is use a Test Driven Development methodology with Claude Sonnet (or recently GPT-5). Moving forward the feature in discrete steps with initial tests and within the red/green loop. I don't see a lot written or discussed about that approach so far, but then reading Martin's article made me realize that the people most proficient with TDD are not really in the Venn Diagram intersection of those wanting to throw themselves wholeheartedly into using LLMs to agent code. The 'super clippy' autocomplete is not the interesting way to use them, it's with multiple agents and prompt techniques at different abstraction levels - that's where you can really cook with gas. Many TDD experts have great pride in the art of code, communicating like a human and holding the abstractions in their head, so we might not get good guidance from the same set of people who helped us before. I think there's a nice green field of 'how to write software' lessons with these tools coming up, with many caution stories and lessons being learnt right now.
It feels like Tdd/llm connection is implied — “and also generate tests”. Thought it’s not cannonical tdd of course. I wonder if it’ll turn the tide towards tech that’s easier to test automatically, like maybe ssr instead of react.
Yep, it's great for generating tests and so much of that is boilerplate that it feels great value. As a super lazy developer it's great as the burden of all that mechanical 'stuff' being spat out is nice. Test code being like baggage feels lighter when it's just churned out as part of the process, as in no guilt just to delete it all when what you want to do changes. That in itself is nice. Plus of course MCP things (Playwright etc) for integration things is great.
But like you said, it was meant more TDD as 'test first' - so a sort of 'prompt-as-spec' that then produces the test/spec code first, and then go iterate on that. The code design itself is different as influenced by how it is prompted to be testable. So rather than go 'prompt -> code' it's more an in-between stage of prompting the test initially and then evolve, making sure the agent is part of the game of only writing testable code and automating the 'gate' of passes before expanding something. 'prompt -> spec -> code' repeat loop until shipped.
The only thing I dislike is what it chooses to test when asked to just "generate tests for X": it often chooses to build those "straitjacket for your code" style tests which aren't actually useful in terms of catching bugs, they just act as "any change now makes this red"
As a simple example, a "buildUrl" style function that put one particular host for prod and a different host for staging (for an "environment" argument) had that argument "tested" by exactly comparing the entire functions return string, encoding all the extra functionality into it (that was tested earlier anyway).
A better output would be to check startsWith(prodHost) or similar, which is what I changed it into, but I'm still trying to work out how to get coding agents to do that in the first or second attempt.
But that's also not surprising: people write those kinds of too-narrow not-useful tests all the time, the codebase I work on is littered with them!
LLMs (Sonnet, Gemini from what I tested) tend to “fix” failing tests by either removing them outright or tweaking the assertions just enough to make them pass. The opposite happens too - sometimes they change the actual logic when what really needs updating is the test.
In short, LLMs often get confused about where the problem lies: the code under test or the test itself. And no amount of context engineering seems to solve that.
I think in part the issue is that the LLM does not have enough context. The difference between a bug in the test or a bug in the implementation is purely based on the requirements which are often not in the source code and stored somewhere else(ticket system, documentation platform).
Without providing the actual feature requirements to the LLM(or the developer) it is impossible to determine which is wrong.
Which is why I think it is also sort of stupid by having the LLM generate tests by just giving it access to the implementation. That is at best testing the implementation as it is, but tests should be based on the requirements.
Oh, absolutely, context matters a lot. But the thing is, they still fail even with solid context.
Before I let an agent touch code, I spell out the issue/feature and have it write two markdown files - strategy.md and progress.md (with the execution order of changes) inside a feat_{id} directory. Once I’m happy with those, I wipe the context and start fresh: feed it the original feature definition + the docs, then tell it to implement by pulling in the right source code context. So by the time any code gets touched, there’s already ~80k tokens in play. And yet, the same confusion frequently happens.
Even if I flat out say “the issue is in the test/logic,”, even if I point out _exactly_ what the issue is, it just apologizes and loops.
At that point I stop it, make it record the failure in the markdown doc, reset context, and let it reload the feature plus the previous agent’s failure. Occasionally that works, but usually once it’s in that state, I have to step in and do it myself.
Not sure. If the Flash image output is $30/M [1] then that's pretty similar to gpt-image-1 costs. So a faster and better model perhaps but not really cheaper?
Since I can't edit, it seems like Flash image is about 23% (4 cents vs 17 cents) of the cost of Openai gpt-image-1, if you're putting an image and prompt in and getting out, say, a 1024x1024 generated image. With the quicker production time that makes it interesting. Expecting Openai to respond at least in terms of pricing, e.g. a flat rate output cap price or something to be comparable.
