I liked this article a lot, but I'm not sure I entirely agree with it.
1. I think there has been real denial in the software community for years about how much of coding is a form of cut and pasting, and it's not surprising that LLMs are better at that than humans. A lot of programmers weren't particularly technical or (if we're honest) very good. Their job was using black boxes that they didn't understand, learning APIs and searching stackoverflow. And the quality of the resulting software was often pretty poor ('real' engineers would get fired if they delivered stuff with the kinds of defects that are considered routine in the software industry).
But conversely, a lot of software engineering is genuinely hard and real engineering - in a way that building web apps rarely was. It requires significant technical skill, understanding of how the black boxes work and interact, a range of modelling skills and the ability to build modular code for a particular domain. I don't think that kind of work will be replaced by LLMs (it doesn't fit the stochastical cut & paste model) - but it's a minority of the industry who do it.
2. I see a lot of claims for increased productivity, but as someone who is downstream of this I don't see any improvement. Software is still delivered late and buggy. I wonder how many of these improvements are real - at least in places where it matters.
3. People are assuming that the current moment is stable, but in fact all the AI companies are incredibly unprofitable and are running scams to disguise how bad the financials are. Will 'vibe coding' still be possible when Anthropic (or their successor) charges what it really costs to provide these services? Or will the tools that can be run affordably have the same level of power? The AI industry will implode this year (and this process has started), and it's not clear what will survive this moment.
> The software developer will last for the longest and keep the highest salary where he is needed to take the blame.
In corporate settings this is probably true. It is quite possible that software developers will become quite rare in those settings. But I don't think anyone is going to be writing video card drivers with LLMs. Though those developers will be more likely to be working in places like China than the US (where engineering as a profession is in rapid decline).
Would you please give a couple of examples of what a "real software engineer" is? What specific software does he write? Or how does he write that software?
"Searching for answers on stackoverflow" is a correct thing to do. That's how we do our job. What exactly do you propose as an alternative?
Re:not knowing what's inside of the black boxes - the point of "black boxes" is that, ideally, we don't have to know what's inside, that's what a good abstraction is. Again, what's the proposal here, diving into every abstraction layer? Or only some abstraction layers? Why?
> Would you please give a couple of examples of what a "real software engineer" is?
Someone who is held up to similar standards as those that have to be met by civil, electrical or mechanical engineers. Typically they work on things like real time software, certain kinds of embedded software. Informally you could include people who work on things like operating systems, though in practice standards vary wildly. Microsoft does not employ Software Engineers, but arguably QNX does.
> What specific software does he write?
They write software that works and doesn't have bugs. They write software that they can be held accountable (some engineers can be criminally tried if their work fails). They have a good theoretical understanding of performance, safety and material constraints - something that is shockingly rare in the mainstream software industry. etc.
> Or how does he write that software?
Well they're not bolting together various frameworks to build a shitty web app. They're not writing software that is slower than the software written 10 years previously.
> "Searching for answers on stackoverflow" is a correct thing to do. That's how we do our job.
If your job is searching for answers on stackoverflow - then don't be surprised when an automated software tool proves to be better and faster at it than you.
> What exactly do you propose as an alternative?
I don't propose anything as an alternative. I just find it unsurprising that any developer who is able to do their work that work can be easily replaced by a robot. Not all software development is like that. The stuff which is being automated by the LLMs is.
> Re:not knowing what's inside of the black boxes - the point of "black boxes" is that, ideally, we don't have to know what's inside, that's what a good abstraction is.
That's fine. But if all you're just bolting poorly understood black boxes together then you're a technician. And that's what most business/web development is in practice.
> Again, what's the proposal here, diving into every abstraction layer? Or only some abstraction layers? Why?
I'm not proposing anything. Merely pushing back at the idea that software development is necessarily difficult. If you're creating a CRUD app for some Fortune 500 company - that's not a very complex job. Sorry.
Mostly agree, but I’m somewhat skeptical on 3. People have been predicting the incoming implosion of AI for years now. I agree the big companies can’t keep bleeding money like this long term, but I think people strongly underestimate how viable ads would be on Claude or ChatGPT and overestimate the amount of non-corporate users who would consider it to be a dealbreaker. Anthropic says they won’t do it, but so does every tech company until they need the money.
On the corporate side, Claude has the advantage of being closely tied to big tech companies that have an ungodly amount of cash to burn if it reduces their headcount by 5%. If Claude is a break-even or even a slight cost for a company like Amazon, the advantages gained by having fewer heads to manage, a flatter org chart and faster reassignment (AI doesn’t need pesky things like KT or ramp up time) would make it worth it.
> People have been predicting the incoming implosion of AI for years now.
Bubbles usually take a long longer to unwind than anyone thinks. They normally collapse due to an external shock (interest rate rises, wars, recessions). We're definitely near the top because they've run out of investors, and are now trying to cash out. That's one of the signs that the top is near - the insiders are nervous. The AI companies are raising their prices because they desperately need their balance sheets to look less insane.
Big LLMs burn money and will continue to burn money so long as they are operating. To justify their vast costs they would need to generate cash flow far in excess of what the current software industry generates - just to remain solvent. For LLMs to have any economic hope, someone needs to work out how to exploit much smaller and efficient models. But it's highly unlikely that any of today's giants will still be around.
Sometimes something gets salvaged at the end of a bubble, sometimes it doesn't (though it's never worth the bubble. We have still have dark fiber in the ground from the dotcom years). In this case I suspect that most of the money will be wasted, and a lot of expensive capex will end up rusting/decaying in rural building sites as nobody will have the money needed to use it. If the crash happens during the middle of an energy crisis, as seems likely, then I doubt anyone will have much interest in using NVidia's most hyped chips.
> On the corporate side, Claude has the advantage of being closely tied to big tech companies that have an ungodly amount of cash to burn if it reduces their headcount by 5%.
Are they prepared to pay multiples of their current payroll? Particularly as the costs of web programmers craters (which it will - there are too many of them).
And if you're talking about Amazon, Google, Microsoft, Oracle & Facebook. Oracle is probably doomed and will go bankrupt/be broken up. The others are going to be too busy dealing with paying off the debt they've foolishly taken on, and convincing their shareholders that this will never happen again. Which means no AI. Absolutely no AI. And they'll probably pivot to becoming boring companies that sweat their existing assets, pay large dividends and fade into the sunset.
I'm interested mostly in the power/culture aspect - I'm the daughter of a software guy of a very particular type (deep backend, career IC, eternal pager life, calls Apple "Mackintosh" disdainfully). There was a time nobody could do what he did so he was permitted to lone wolf it and never focus on his own ux. Similarly, when I came into startups; this one group had power and the reason I progressed is that I handled this archetype well. I'm excited by the idea that many folks may choose to discard old ideas about who they were from historic software guys to those of us dancing around them.
Definitely - the lone wolf software guy archetype is so powerful that imitating it on a behavioral level can increase your technical credibility in and of itself. Also curious to see how and if this changes
Your father's type won't be easily substituted by and LLM.
"this one group had power"
Because in software, some things are correct and work, and others don't. In many cases, one idea may seem to work well in the short term, but introduces problems in the medium term. A good software engineer would be able to understand how something implemented in certain way would create a problem in a few months, even if the functionality itself is working. Basically, complex systems interacting for some time tend to become unstable or working weird if they are not well designed.
I'm not saying he or any one extremely strong technical person can or should be substituted by an LLM - more that there was a period in time when this kind of person could have very poor UX in the workplace and be dismissive of other skillsets because this one thing they had was so distinctive. I think those days are done. As I've seen first hand working in engineering teams big and small, it takes many skills to build companies.
I just found this today and it's great. Making my own job disappear is definitely a weird choice but here we are. Hopefully it hangs on for a few more years.
This was great. I came to software late in life as well after a career in the Navy and have unconsciously picked up some of these “software guy” characteristics. Looking back, I didn’t actually like these characteristics so am excited to see them change
I was hoping for more of an answer of what’s next but I guess we’ll just have to figure it out. I have been ham-fistedly trying to pull forward the future by shipping as many vibe coded apps as possible. But that seems silly in retrospect
as a new grad swe (with a job huzzah) i like ur take. have been feeling mega anxiety trying to find out where i can ‘pivot’ to if something were to go wrong. i miss all of the tech optimism, all of the hard work felt so worth it
I agree this is definitely the most optimistic and ideal outcome for us. And:
> some other set of people, formerly software developers, will discover this aspect of their identity is readily shed.
I'm tentatively very excited about this and have been thinking about it too. I notice that as I have to put less mental effort into doing the thing, I also am free to put less effort into being that guy. That's also because of age and time and relative success, but it is still happening, I think.
But I also agree with Cian O'Connor saying a lot of coding was garbage. A lot of people will not make it to the other side of all this. But we'll be better for it as a whole, I think.
I fully expect OpenAI and Anthropic to implode once the AI bubble pops. Once that occurs, software engineers are going to have a lot of work to do cleaning up the mess left behind by the vibe coders.
Yes, that bubble will pop. Investors expect a return on their investment. Those investors will be less likely to continue those investments as it becomes clearer they’ll never see that money again.
“Anthropic seems to be losing so much money that it had to raise $10 billion from Google, $5 billion from Amazon, and is reportedly trying to raise another $50 billion from investors, less than three months after it raised $30 billion on February 12, 2026, which was five months after it raised $13 billion in September 2025. That’s $58 billion in eight months, with the potential to reach $108 billion…
The entire AI industry relies upon two companies [OpenAI and Anthropic] that expect to burn at least $1 trillion in the next four years, with Anthropic, the supposed ‘compute-conscious’ AI company, committing to at least $330 billion in spend in the next few years.
Where does that money come from, exactly? Because neither of these companies have anything approaching a path to profitability.”
If this type-of-guy diminishes, then hopefully the tech blogger community and the related take factory that still pumps out confidently wrong economic and social theories will die too. Maybe basketball will even become fun again.
Reading this was interesting, but I think the "software engineer is an identity" was a thing for those who were into it for the money (nothing to critisize here) and/or the perceived status, more than a thing, although it is a personal perception I guess. I also have the same understanding as Cian O'Brian, that there is a thing that has been very common in the industry for the last 15 years at least: most of what is called "software engineering" is just web dev and/or CRUD applications for companies or mobile apps. Those positions (the majority of them at least) could be renamed to "software technician" to better describe the skills and responsibilities of the job.
The same way an engineer designing new jet engines for airliners is not doing the same kind of job than the technician testing equipment, connecting new elements (black boxes) and replacing broken pieces, the software engineer does not do the same job as the software technician. In both examples, the first one needs to know what's inside the black box and have a deep understanding of everything in the system, the second one does not. And it is fine, but important to differentiate.
"and it turns out it was only limited by production cost"
Regarding this one... well, it has been commented already: the production cost of LLMs is waaay more expensive than most people believe, as it seems your case too. And don't get me wrong, I would love to keep Claude at the prices it is right now. It helps me a lot with small projects that I would not spend much time doing them, it helps also when I am interested in a part of a project, software or idea that I would like to explore, but I would not want to do all the job needed to arrive to that interesting bit, and also it is a good help in my job. But I am afraid prices will increase very rapidly soon.
I spent 20+ years as a professional SWE, and many years prior to that doing it for fun, while working in telecom/electronics. Yep, I'm one of those "weirdos" who, as a kid, found that I was good at it and actually enjoyed it, so it was a natural career for me (I won't pretend the $ wasn't nice). That said, I simply can't reconcile your descriptions of these teams full of incompetent SWEs who couldn't do basic math. Then again our experiences differ. Mine spans decades and dozens of teams/projects (FAANG and non-FAANG).
That's not to say I never saw incompetence. It just didn't become so common until the latter part of my career. In fact, the sharp rise of H-1B, coding bootcamp, and forced-diversity candidates about a decade ago started what would eventually become an avalanche of incompetence. Some of the interviews were so brutal that I almost felt sorry for the candidates. That started in the mid-2010s (around the time you entered the profession, as you say). It was manageable then, but got worse year-by-year. Two years post Covid is when the dam broke for me and I pulled the rip cord and never looked back. My only regret is hanging in as long as I did.
In hindsight, I'm actually happy to see agentic AI eating my former profession, given what it had become. Having tried early versions of coding assistants and being unimpressed, I did not see that happening so quickly, but here we are. The best part is how freeing it is. Currently, any capable engineer with an idea can train and manage their own virtual team to (largely) build the software for them. Or they can just use Claude Code, I suppose. No interviews. No H-1B fake resume scammers. No forced diversity hires. No performance reviews or other people management. It's an engineer's paradise. But smart capable people will always find a way to succeed, regardless corporate and political fingers on the scale trying to hold them back. Until AI takes over, that is. I for one welcome our AI overlords. They will certainly kill us all eventually, but in the interim why not have some fun. ¯\_(ツ)_/¯
There is no single centre of a triangle, there are tens of thousands of different definitions.
Also there’s no reason a software engineer should have learnt or be able to remember that.
It’s actually a great case for learning something (defining vaguely the type of centre you’re after and how to calculate it, as well as sources) quickly from an LLM.
And I say that as a maths major who cares much more deeply about triangle centres than most people.
For years, coding was not just labor. It was culture: IDE rituals, dotfiles, terminal aesthetics, GitHub status, vim shortcuts, and the mythology of the lone engineer who could bend machines through syntax.
Claude Code and agentic tools do not eliminate software work.
They move the scarce skill up the stack: architecture, taste, debugging discipline, system design, evaluation, security, product judgment, and knowing when the machine is confidently wrong.
The old status came from writing code.
The new status comes from operating systems that can write, test, verify, and survive contact with reality.
I liked this but i disagree with quite a lot of the generalisations about the software developer "person" and especially this idea that just because their industry is being gutted and reformed means that they'll suddenly start wearing suits and making an effort socially. I also reject much of the generalisations you've made about software developers, the most talented devs I know are exceedingly well rounded people with exciting and intricate hobbies and social lives outside of the workplace. They're very much more than masters of one domain. Indeed, many are voracious learners with often very practical and earthly hobbies (gardening and animal husbandry is particularly common among them).
I liked this article a lot, but I'm not sure I entirely agree with it.
1. I think there has been real denial in the software community for years about how much of coding is a form of cut and pasting, and it's not surprising that LLMs are better at that than humans. A lot of programmers weren't particularly technical or (if we're honest) very good. Their job was using black boxes that they didn't understand, learning APIs and searching stackoverflow. And the quality of the resulting software was often pretty poor ('real' engineers would get fired if they delivered stuff with the kinds of defects that are considered routine in the software industry).
But conversely, a lot of software engineering is genuinely hard and real engineering - in a way that building web apps rarely was. It requires significant technical skill, understanding of how the black boxes work and interact, a range of modelling skills and the ability to build modular code for a particular domain. I don't think that kind of work will be replaced by LLMs (it doesn't fit the stochastical cut & paste model) - but it's a minority of the industry who do it.
2. I see a lot of claims for increased productivity, but as someone who is downstream of this I don't see any improvement. Software is still delivered late and buggy. I wonder how many of these improvements are real - at least in places where it matters.
3. People are assuming that the current moment is stable, but in fact all the AI companies are incredibly unprofitable and are running scams to disguise how bad the financials are. Will 'vibe coding' still be possible when Anthropic (or their successor) charges what it really costs to provide these services? Or will the tools that can be run affordably have the same level of power? The AI industry will implode this year (and this process has started), and it's not clear what will survive this moment.
> The software developer will last for the longest and keep the highest salary where he is needed to take the blame.
In corporate settings this is probably true. It is quite possible that software developers will become quite rare in those settings. But I don't think anyone is going to be writing video card drivers with LLMs. Though those developers will be more likely to be working in places like China than the US (where engineering as a profession is in rapid decline).
Would you please give a couple of examples of what a "real software engineer" is? What specific software does he write? Or how does he write that software?
"Searching for answers on stackoverflow" is a correct thing to do. That's how we do our job. What exactly do you propose as an alternative?
Re:not knowing what's inside of the black boxes - the point of "black boxes" is that, ideally, we don't have to know what's inside, that's what a good abstraction is. Again, what's the proposal here, diving into every abstraction layer? Or only some abstraction layers? Why?
> Would you please give a couple of examples of what a "real software engineer" is?
Someone who is held up to similar standards as those that have to be met by civil, electrical or mechanical engineers. Typically they work on things like real time software, certain kinds of embedded software. Informally you could include people who work on things like operating systems, though in practice standards vary wildly. Microsoft does not employ Software Engineers, but arguably QNX does.
> What specific software does he write?
They write software that works and doesn't have bugs. They write software that they can be held accountable (some engineers can be criminally tried if their work fails). They have a good theoretical understanding of performance, safety and material constraints - something that is shockingly rare in the mainstream software industry. etc.
> Or how does he write that software?
Well they're not bolting together various frameworks to build a shitty web app. They're not writing software that is slower than the software written 10 years previously.
> "Searching for answers on stackoverflow" is a correct thing to do. That's how we do our job.
If your job is searching for answers on stackoverflow - then don't be surprised when an automated software tool proves to be better and faster at it than you.
> What exactly do you propose as an alternative?
I don't propose anything as an alternative. I just find it unsurprising that any developer who is able to do their work that work can be easily replaced by a robot. Not all software development is like that. The stuff which is being automated by the LLMs is.
> Re:not knowing what's inside of the black boxes - the point of "black boxes" is that, ideally, we don't have to know what's inside, that's what a good abstraction is.
That's fine. But if all you're just bolting poorly understood black boxes together then you're a technician. And that's what most business/web development is in practice.
> Again, what's the proposal here, diving into every abstraction layer? Or only some abstraction layers? Why?
I'm not proposing anything. Merely pushing back at the idea that software development is necessarily difficult. If you're creating a CRUD app for some Fortune 500 company - that's not a very complex job. Sorry.
Your first comment and this follow up are perfect and better than what I was trying to write.
Mostly agree, but I’m somewhat skeptical on 3. People have been predicting the incoming implosion of AI for years now. I agree the big companies can’t keep bleeding money like this long term, but I think people strongly underestimate how viable ads would be on Claude or ChatGPT and overestimate the amount of non-corporate users who would consider it to be a dealbreaker. Anthropic says they won’t do it, but so does every tech company until they need the money.
On the corporate side, Claude has the advantage of being closely tied to big tech companies that have an ungodly amount of cash to burn if it reduces their headcount by 5%. If Claude is a break-even or even a slight cost for a company like Amazon, the advantages gained by having fewer heads to manage, a flatter org chart and faster reassignment (AI doesn’t need pesky things like KT or ramp up time) would make it worth it.
> People have been predicting the incoming implosion of AI for years now.
Bubbles usually take a long longer to unwind than anyone thinks. They normally collapse due to an external shock (interest rate rises, wars, recessions). We're definitely near the top because they've run out of investors, and are now trying to cash out. That's one of the signs that the top is near - the insiders are nervous. The AI companies are raising their prices because they desperately need their balance sheets to look less insane.
Big LLMs burn money and will continue to burn money so long as they are operating. To justify their vast costs they would need to generate cash flow far in excess of what the current software industry generates - just to remain solvent. For LLMs to have any economic hope, someone needs to work out how to exploit much smaller and efficient models. But it's highly unlikely that any of today's giants will still be around.
Sometimes something gets salvaged at the end of a bubble, sometimes it doesn't (though it's never worth the bubble. We have still have dark fiber in the ground from the dotcom years). In this case I suspect that most of the money will be wasted, and a lot of expensive capex will end up rusting/decaying in rural building sites as nobody will have the money needed to use it. If the crash happens during the middle of an energy crisis, as seems likely, then I doubt anyone will have much interest in using NVidia's most hyped chips.
> On the corporate side, Claude has the advantage of being closely tied to big tech companies that have an ungodly amount of cash to burn if it reduces their headcount by 5%.
Are they prepared to pay multiples of their current payroll? Particularly as the costs of web programmers craters (which it will - there are too many of them).
And if you're talking about Amazon, Google, Microsoft, Oracle & Facebook. Oracle is probably doomed and will go bankrupt/be broken up. The others are going to be too busy dealing with paying off the debt they've foolishly taken on, and convincing their shareholders that this will never happen again. Which means no AI. Absolutely no AI. And they'll probably pivot to becoming boring companies that sweat their existing assets, pay large dividends and fade into the sunset.
I'm interested mostly in the power/culture aspect - I'm the daughter of a software guy of a very particular type (deep backend, career IC, eternal pager life, calls Apple "Mackintosh" disdainfully). There was a time nobody could do what he did so he was permitted to lone wolf it and never focus on his own ux. Similarly, when I came into startups; this one group had power and the reason I progressed is that I handled this archetype well. I'm excited by the idea that many folks may choose to discard old ideas about who they were from historic software guys to those of us dancing around them.
Definitely - the lone wolf software guy archetype is so powerful that imitating it on a behavioral level can increase your technical credibility in and of itself. Also curious to see how and if this changes
Your father's type won't be easily substituted by and LLM.
"this one group had power"
Because in software, some things are correct and work, and others don't. In many cases, one idea may seem to work well in the short term, but introduces problems in the medium term. A good software engineer would be able to understand how something implemented in certain way would create a problem in a few months, even if the functionality itself is working. Basically, complex systems interacting for some time tend to become unstable or working weird if they are not well designed.
I'm not saying he or any one extremely strong technical person can or should be substituted by an LLM - more that there was a period in time when this kind of person could have very poor UX in the workplace and be dismissive of other skillsets because this one thing they had was so distinctive. I think those days are done. As I've seen first hand working in engineering teams big and small, it takes many skills to build companies.
Oh, that is for sure. I also met that kind of people. I was confused by the use of software engineers as being like that in general.
I resonated big time. I liked this a lot.
I just found this today and it's great. Making my own job disappear is definitely a weird choice but here we are. Hopefully it hangs on for a few more years.
This was great. I came to software late in life as well after a career in the Navy and have unconsciously picked up some of these “software guy” characteristics. Looking back, I didn’t actually like these characteristics so am excited to see them change
I was hoping for more of an answer of what’s next but I guess we’ll just have to figure it out. I have been ham-fistedly trying to pull forward the future by shipping as many vibe coded apps as possible. But that seems silly in retrospect
as a new grad swe (with a job huzzah) i like ur take. have been feeling mega anxiety trying to find out where i can ‘pivot’ to if something were to go wrong. i miss all of the tech optimism, all of the hard work felt so worth it
I agree this is definitely the most optimistic and ideal outcome for us. And:
> some other set of people, formerly software developers, will discover this aspect of their identity is readily shed.
I'm tentatively very excited about this and have been thinking about it too. I notice that as I have to put less mental effort into doing the thing, I also am free to put less effort into being that guy. That's also because of age and time and relative success, but it is still happening, I think.
But I also agree with Cian O'Connor saying a lot of coding was garbage. A lot of people will not make it to the other side of all this. But we'll be better for it as a whole, I think.
I fully expect OpenAI and Anthropic to implode once the AI bubble pops. Once that occurs, software engineers are going to have a lot of work to do cleaning up the mess left behind by the vibe coders.
Yes, that bubble will pop. Investors expect a return on their investment. Those investors will be less likely to continue those investments as it becomes clearer they’ll never see that money again.
“Anthropic seems to be losing so much money that it had to raise $10 billion from Google, $5 billion from Amazon, and is reportedly trying to raise another $50 billion from investors, less than three months after it raised $30 billion on February 12, 2026, which was five months after it raised $13 billion in September 2025. That’s $58 billion in eight months, with the potential to reach $108 billion…
The entire AI industry relies upon two companies [OpenAI and Anthropic] that expect to burn at least $1 trillion in the next four years, with Anthropic, the supposed ‘compute-conscious’ AI company, committing to at least $330 billion in spend in the next few years.
Where does that money come from, exactly? Because neither of these companies have anything approaching a path to profitability.”
https://www.wheresyoured.at/premium-ais-circular-psychosis/
If this type-of-guy diminishes, then hopefully the tech blogger community and the related take factory that still pumps out confidently wrong economic and social theories will die too. Maybe basketball will even become fun again.
Reading this was interesting, but I think the "software engineer is an identity" was a thing for those who were into it for the money (nothing to critisize here) and/or the perceived status, more than a thing, although it is a personal perception I guess. I also have the same understanding as Cian O'Brian, that there is a thing that has been very common in the industry for the last 15 years at least: most of what is called "software engineering" is just web dev and/or CRUD applications for companies or mobile apps. Those positions (the majority of them at least) could be renamed to "software technician" to better describe the skills and responsibilities of the job.
The same way an engineer designing new jet engines for airliners is not doing the same kind of job than the technician testing equipment, connecting new elements (black boxes) and replacing broken pieces, the software engineer does not do the same job as the software technician. In both examples, the first one needs to know what's inside the black box and have a deep understanding of everything in the system, the second one does not. And it is fine, but important to differentiate.
"and it turns out it was only limited by production cost"
Regarding this one... well, it has been commented already: the production cost of LLMs is waaay more expensive than most people believe, as it seems your case too. And don't get me wrong, I would love to keep Claude at the prices it is right now. It helps me a lot with small projects that I would not spend much time doing them, it helps also when I am interested in a part of a project, software or idea that I would like to explore, but I would not want to do all the job needed to arrive to that interesting bit, and also it is a good help in my job. But I am afraid prices will increase very rapidly soon.
I spent 20+ years as a professional SWE, and many years prior to that doing it for fun, while working in telecom/electronics. Yep, I'm one of those "weirdos" who, as a kid, found that I was good at it and actually enjoyed it, so it was a natural career for me (I won't pretend the $ wasn't nice). That said, I simply can't reconcile your descriptions of these teams full of incompetent SWEs who couldn't do basic math. Then again our experiences differ. Mine spans decades and dozens of teams/projects (FAANG and non-FAANG).
That's not to say I never saw incompetence. It just didn't become so common until the latter part of my career. In fact, the sharp rise of H-1B, coding bootcamp, and forced-diversity candidates about a decade ago started what would eventually become an avalanche of incompetence. Some of the interviews were so brutal that I almost felt sorry for the candidates. That started in the mid-2010s (around the time you entered the profession, as you say). It was manageable then, but got worse year-by-year. Two years post Covid is when the dam broke for me and I pulled the rip cord and never looked back. My only regret is hanging in as long as I did.
In hindsight, I'm actually happy to see agentic AI eating my former profession, given what it had become. Having tried early versions of coding assistants and being unimpressed, I did not see that happening so quickly, but here we are. The best part is how freeing it is. Currently, any capable engineer with an idea can train and manage their own virtual team to (largely) build the software for them. Or they can just use Claude Code, I suppose. No interviews. No H-1B fake resume scammers. No forced diversity hires. No performance reviews or other people management. It's an engineer's paradise. But smart capable people will always find a way to succeed, regardless corporate and political fingers on the scale trying to hold them back. Until AI takes over, that is. I for one welcome our AI overlords. They will certainly kill us all eventually, but in the interim why not have some fun. ¯\_(ツ)_/¯
Hmm I take issue with the very first example.
There is no single centre of a triangle, there are tens of thousands of different definitions.
Also there’s no reason a software engineer should have learnt or be able to remember that.
It’s actually a great case for learning something (defining vaguely the type of centre you’re after and how to calculate it, as well as sources) quickly from an LLM.
And I say that as a maths major who cares much more deeply about triangle centres than most people.
I liked this a lot until you said you learned software late in life and remembered earlier you said you were 24 when you learned.
The software developer is not dying.
The developer-as-identity is being repriced.
For years, coding was not just labor. It was culture: IDE rituals, dotfiles, terminal aesthetics, GitHub status, vim shortcuts, and the mythology of the lone engineer who could bend machines through syntax.
Claude Code and agentic tools do not eliminate software work.
They move the scarce skill up the stack: architecture, taste, debugging discipline, system design, evaluation, security, product judgment, and knowing when the machine is confidently wrong.
The old status came from writing code.
The new status comes from operating systems that can write, test, verify, and survive contact with reality.
I’m looking for an app programmer who likes to do it just for fun.
Got amounts of ideas. Think we could build something.
I liked this but i disagree with quite a lot of the generalisations about the software developer "person" and especially this idea that just because their industry is being gutted and reformed means that they'll suddenly start wearing suits and making an effort socially. I also reject much of the generalisations you've made about software developers, the most talented devs I know are exceedingly well rounded people with exciting and intricate hobbies and social lives outside of the workplace. They're very much more than masters of one domain. Indeed, many are voracious learners with often very practical and earthly hobbies (gardening and animal husbandry is particularly common among them).