This will take a while for most people to internalize, but now with GPT-6-Astra we are basically in the age where as a developer you control agents to write code for you. This will be further cemented by the next generation of models.
I fear that the rest of society is not at all keeping up with the pace of development, and we're quite soon going to face a situation where most white collar work will be done by agents. What this might do for employment is worrying.
> but now with GPT-6-Astra we are basically in the age where as a developer you control agents to write code for you
Given I've kind of felt like that since using GPT-4.1 in Codex, I'm wondering if this is not a "Now we all can" boolean state, but a threshold that keeps moving downwards, enabling more and more people to build better (or "more" maybe, not necessarily "better").
You're still saying "as a developer you control" though, so one would assume it'd be the same for rest of white collar work though. And since we're all obviously familiar with Jevons paradox (https://en.wikipedia.org/wiki/Jevons_paradox), it might not lead to a downturn in employment at all. But on the other hand, maybe all it takes is a "developer" regardless of sector, so everyone will end up being a d̴e̴v̴e̴l̴o̴p̴e̴r̴ agent hand-holder.
I guess should low-key be happy that even expenses of 50 dollars per month for AI are being seen at ludicrous. Combined with our Microsoft only policy (i.e. ghcp), there's still hardly any serious ai use going on. Ah yes EU high tech companies...
Do you work in a well-trodden set of problem spaces? There is still so much room for humans to be writing code. Especially in anything moderately creative.
Yeah, the industry will quickly shed all of the people who needed it to be one way and will grow with all the people who are capable of it being a different way.
It's hard to take the title seriously when the first paragraph ends with "It’s hard to know right now if the investment will pay dividends or not.".
I too think that probably the mechanical action of putting fingers against a keyboard to inspect, understand, debug and build software might effectively end up like writing punch cards is today. But it still irks me if we'd try to shoehorn in getting there instead of actually looking at this taking all the sides into account, including if it's actually worth it given what it costs today or will cost tomorrow.
Before commenting, it's worth reading through the entire thing (big surprise), but especially this:
> Software development IS NOT CODING
The author is talking about a very specific part of software development, not all of it, as some might understand by "coding".
Biggest benefit of all this getting the control of hardware back to consumers. Software has been the soul of the computers which only knowledgeable enough could put into hardware for a long time.
Now (almost) anyone who owns a programmable device also controls it the way they want, as if they actually own it.
I am stretching a bit here but we are almost there now already.
I wish these models start writing super optimal code that can squeeze very last bit of the chips so that we can get rid of Electron and similar bloatware.
I think it should be doable now. Someone might have tried this already. I do know people connecting AI to TVs to cleanup all the bloat automatically and it works. Saw a few posts on twitter.
Worked on my first borderline-computer-sciency not 100% trivial task in a long while today: Not sending one SNS message per event, but batching them on the client side, sending out as batched message if it went beyond a certain size or beyond a certain timeout (-> limit event delay caused by waiting for batch to fill up.) (this will save A LOT of money)
I wrote a nice spec for it, gave it to claude fable and it completely shit the bed. I ended up writing most of it myself and let claude iron out the bugs and kinks. Ultimately it was just 150 lines and far easier to follow and much more sensible than the fully automated claude output. I don't understand why AI couldn't build this.
Apparently I am still not living in the dark-software-factory parallel world many other people live in. Am I already part of the permanent AI underclass? Lapped and too far behind to catch up?
I think a lot of the people who say coding is dead either have low standards or or are low skilled. If you still think the quality of the codebase matters then agents absolutely cannot do the job of a human yet. The only way to know this is to review their code and understand how bone headed the decisions made by agents are. Sure, it may solve the problem and add unit tests, but it's absolutely creating junk while doing so.
I actually think everyone should still code a little. Maybe 30-60 minutes per day. While the majority of code is generated by agents, I've found that with even a little coding I catch stuff that I miss by just doing code reviews of agentic-written code. But yeah, the majority of my day isn't coding now.
Part of software development is architectural design, database design, etc. That goes hand-in-hand with coding -- you come up with the design then implement that design.
Part of software development is structuring the code to make it readable and maintainable.
Part of software development is writing suitable tests to ensure that the code does what it is required to by the design and specification.
Part of software development is ensuring that the code meets things like accessible and security guidelines.
Part of software development is choosing suitable technologies and languages.
It's been tomorrow for an awfully long time now. Does it feel like the quality of software has got better or worse in the last few years? Is larger quantities of worse software really want we want to optimise for as an industry?
Entry level coding has changed in that it can be replaced by AI. But this is a preference, not a rule or law. However, AI is still many years or even decades away from writing good complicated well thought out software. This is software that big F500 companies typically spend a lot of money on. This software allows F500 employees to outsource their work to AI since the hard complicated stuff is taken care of. One way to think about this is AI writing. Everyone agrees AI prose is shallow, repetitive, lacks creative structure, and can be slop like. By no means is writing dead. In fact, it’s the opposite, good writers have now become more valuable. This is something the author mentions.
>I’m a software engineer at JPMorgan Chase. I have more than 15 years of experience in building software and I used to love it
Big woop :) I have 50 years as a professional programmer and all AI will do is create more programming jobs. Many times I have seen the latest and greatest thing that will eliminate programming. You know what all of those did ? Increase demand for programmers. The exact opposite of what management wanted. AI, same will happen once the shine wears off.
My concern is not loosing my job. I agree with your point. It's likely true that demand will increase. I'm just mourning the loss of a craft I enjoyed.
When I say "get over it" I'm talking to myself.
I remember when C was a "high level language" and model driven development would give us IDEs where coding would be on a "higher abstraction level" and code would be just "generated" ad hoc and disposable.
Some people never internalize thermodynamics. You can't have irreversible processes, i.e. systems evolution, without generating excess entropy. There is no technology that will make entropy vanish. There will be a trail of entropy exhaust all the way from the highest level of abstraction you can imagine down to the bottom.
No amount of (big) data or GPU farms will ever rid us of shoveling entropy out.
Yet people keep promising this entropy sink that will magically give us all value no cost like magic.
> Model capabilities are evolving so fast that what was true last month may not be true today
I dunno. I think the harnesses are getting a lot better but older models still do the bread and butter just as good.
> Recently the head of Claude Code, Boris Cherny, said to delete all your AGENTS.md Claude.md files and just see what the models can do nowadays
It was pretty obvious early on (which is only like a year ago) that you shouldn’t have much outside of things the model/person has no way of knowing inside your agents.md. Basically just a pointer to your docs. That OP didn’t know this as well as they listen to number one shovel salesman Boris makes them seem unserious.
> My process for learning a new domain or technology used to be modeling it in code: build to learn
This is still the process. My guess is OP is actually just an average / below average developer, so AI causes this reaction. I also have this reaction anytime I use AI in area I’m a novice in. You still need to learn domains deeply to unlock the ability to use AI effectively. Domain dependent, but if you’re not seeing AI make mistakes consistently you need to learn more about how the tech you’re using works. You’re missing things.
The world is still full of ambiguity and tradeoffs that have no objective correct answer. I don’t think AI or humans have solved all these cases yet.
Headlines like this do more harm than good. I sometimes wonder if those are paid marketing or people actually write them on their own.
He stated that "software development IS NOT CODING", but coding is part of it. Not all, but part. If you outsource that knowledge you will be forever trapped in the walled garden that is (and will get worse) those LLM services that code for you. If you are senior developer you will know when and how to use tools to help you, but people starting in the industry will get the wrong idea.
All that wave of 'X is over because a new LLM is out' is discouraging for anybody new searching for that field and irresponsible. The cynic in me tends to believe this is on purpose to build a vacuum on such markets and fill it with mediocre work.
I see all these comments here dunking on the article, but I have to say, I basically completely agree with all the points the author is making. I think there are different kinds of software engineers, and the author's point of view resonates with those of us that are mostly concerned with building tools, rather than products. At the end of the day, I need the tool to work so that it can be used to build other things. In that sense, coding is dead for me. Anything more than trivial few dozen line changes to the codebase is going to be done by claude. It will figure out the hidden dependencies and gotchas more carefully than I will on a given time budget. And for me, unit tests have always been an afterthought, so the fact that the agent will also add unit tests for its modifications means it's strictly more disciplined than I ever was. I've been at this for 25+ years, and had the fortune to be able to largely be a "cowboy coder" but those days are long gone.
I still love writing code myself, but as I use LLMs more, I'm realizing it's not really the coding I actually loved, it was the software architecture and up front design work. Once the structure and plans are laid out, writing the code is just mechanical, and I'm going to reach for an agent if I can. Of course, these days, I find myself asking claude for advice on architecture as well, but that is much less a deferral of responsibilities than asking for an audit or survey.
Nothing. Nothing prevents me from growing my own food either but I would be in poverty if I did. Gardening/coding is fun but I don't think it's a good investment of my time, given the tools/systems I have at my disposal.
Maybe I'm wrong and I will not miss it. I feel like something is being lost though. A bit of grief is appropriate.
If this is true, why don't I see a flood of high quality AI-written software? All I'm seeing is a bunch of vibe-coded shit that's basically of no use to anyone.
For now, IMO, high quality software still requires great engineers. Not necessarily spending their time writing code.
There is still a shortage of good engineers (maybe there will always be)
I call BS on that, it's been 6 months since Opus 4.6 and we still do not see a revolutionary software release or a major increase in quality. At least we see some security bugs being finally fixed.
Maxion | 14 hours ago
I fear that the rest of society is not at all keeping up with the pace of development, and we're quite soon going to face a situation where most white collar work will be done by agents. What this might do for employment is worrying.
embedding-shape | 13 hours ago
Given I've kind of felt like that since using GPT-4.1 in Codex, I'm wondering if this is not a "Now we all can" boolean state, but a threshold that keeps moving downwards, enabling more and more people to build better (or "more" maybe, not necessarily "better").
You're still saying "as a developer you control" though, so one would assume it'd be the same for rest of white collar work though. And since we're all obviously familiar with Jevons paradox (https://en.wikipedia.org/wiki/Jevons_paradox), it might not lead to a downturn in employment at all. But on the other hand, maybe all it takes is a "developer" regardless of sector, so everyone will end up being a d̴e̴v̴e̴l̴o̴p̴e̴r̴ agent hand-holder.
on_the_train | 12 hours ago
qmmmur | 11 hours ago
lousken | 10 hours ago
spicyusername | 13 hours ago
So the wheel of the market turns.
intended | 13 hours ago
and
> There are good developers, mediocre developers, and bad developers, and the good ones are still hard to find.
move in opposite directions. Friction is the source of learning, which creates good developers.
embedding-shape | 13 hours ago
I too think that probably the mechanical action of putting fingers against a keyboard to inspect, understand, debug and build software might effectively end up like writing punch cards is today. But it still irks me if we'd try to shoehorn in getting there instead of actually looking at this taking all the sides into account, including if it's actually worth it given what it costs today or will cost tomorrow.
Before commenting, it's worth reading through the entire thing (big surprise), but especially this:
> Software development IS NOT CODING
The author is talking about a very specific part of software development, not all of it, as some might understand by "coding".
smusamashah | 13 hours ago
Now (almost) anyone who owns a programmable device also controls it the way they want, as if they actually own it.
I am stretching a bit here but we are almost there now already.
I wish these models start writing super optimal code that can squeeze very last bit of the chips so that we can get rid of Electron and similar bloatware.
xnx | 13 hours ago
smusamashah | 13 hours ago
Traubenfuchs | 13 hours ago
I wrote a nice spec for it, gave it to claude fable and it completely shit the bed. I ended up writing most of it myself and let claude iron out the bugs and kinks. Ultimately it was just 150 lines and far easier to follow and much more sensible than the fully automated claude output. I don't understand why AI couldn't build this.
Apparently I am still not living in the dark-software-factory parallel world many other people live in. Am I already part of the permanent AI underclass? Lapped and too far behind to catch up?
untrust | 12 hours ago
mattm | 13 hours ago
VCFundedGenYer | 13 hours ago
Coding will never be "over" - those saying it aren't true coders in the first place.
embedding-shape | 13 hours ago
rhdunn | 13 hours ago
Part of software development is structuring the code to make it readable and maintainable.
Part of software development is writing suitable tests to ensure that the code does what it is required to by the design and specification.
Part of software development is ensuring that the code meets things like accessible and security guidelines.
Part of software development is choosing suitable technologies and languages.
kige | 12 hours ago
ozlikethewizard | 5 hours ago
KevinMS | 13 hours ago
cheney_2004 | 13 hours ago
jmclnx | 13 hours ago
Big woop :) I have 50 years as a professional programmer and all AI will do is create more programming jobs. Many times I have seen the latest and greatest thing that will eliminate programming. You know what all of those did ? Increase demand for programmers. The exact opposite of what management wanted. AI, same will happen once the shine wears off.
[OP] pitermarx | 12 hours ago
ozlikethewizard | 5 hours ago
huurtehoog | 2 hours ago
I remember when C was a "high level language" and model driven development would give us IDEs where coding would be on a "higher abstraction level" and code would be just "generated" ad hoc and disposable.
Some people never internalize thermodynamics. You can't have irreversible processes, i.e. systems evolution, without generating excess entropy. There is no technology that will make entropy vanish. There will be a trail of entropy exhaust all the way from the highest level of abstraction you can imagine down to the bottom.
No amount of (big) data or GPU farms will ever rid us of shoveling entropy out.
Yet people keep promising this entropy sink that will magically give us all value no cost like magic.
skjsbzbz | 13 hours ago
I dunno. I think the harnesses are getting a lot better but older models still do the bread and butter just as good.
> Recently the head of Claude Code, Boris Cherny, said to delete all your AGENTS.md Claude.md files and just see what the models can do nowadays
It was pretty obvious early on (which is only like a year ago) that you shouldn’t have much outside of things the model/person has no way of knowing inside your agents.md. Basically just a pointer to your docs. That OP didn’t know this as well as they listen to number one shovel salesman Boris makes them seem unserious.
> My process for learning a new domain or technology used to be modeling it in code: build to learn
This is still the process. My guess is OP is actually just an average / below average developer, so AI causes this reaction. I also have this reaction anytime I use AI in area I’m a novice in. You still need to learn domains deeply to unlock the ability to use AI effectively. Domain dependent, but if you’re not seeing AI make mistakes consistently you need to learn more about how the tech you’re using works. You’re missing things.
The world is still full of ambiguity and tradeoffs that have no objective correct answer. I don’t think AI or humans have solved all these cases yet.
harrisoned | 13 hours ago
He stated that "software development IS NOT CODING", but coding is part of it. Not all, but part. If you outsource that knowledge you will be forever trapped in the walled garden that is (and will get worse) those LLM services that code for you. If you are senior developer you will know when and how to use tools to help you, but people starting in the industry will get the wrong idea.
All that wave of 'X is over because a new LLM is out' is discouraging for anybody new searching for that field and irresponsible. The cynic in me tends to believe this is on purpose to build a vacuum on such markets and fill it with mediocre work.
CyLith | 12 hours ago
I still love writing code myself, but as I use LLMs more, I'm realizing it's not really the coding I actually loved, it was the software architecture and up front design work. Once the structure and plans are laid out, writing the code is just mechanical, and I'm going to reach for an agent if I can. Of course, these days, I find myself asking claude for advice on architecture as well, but that is much less a deferral of responsibilities than asking for an audit or survey.
[OP] pitermarx | 12 hours ago
Martin_Silenus | 12 hours ago
When honesty goes out the window...
[OP] pitermarx | 12 hours ago
Martin_Silenus | 11 hours ago
[OP] pitermarx | 11 hours ago
Maybe I'm wrong and I will not miss it. I feel like something is being lost though. A bit of grief is appropriate.
ndsipa_pomu | 11 hours ago
[OP] pitermarx | 11 hours ago
lousken | 10 hours ago