Ask HN: The death of software development as a job?
34 comments
Actually writing code was never the difficult part for the majority of software created. It required skill yes, but the really hard part was figuring out what to implement in the first place.
Which features should the software have, how should they function and interact, which tradeoffs to make given the limitations. Stuff like that.
People who were good at those things but lacked training or capabilities to actually write functioning code can now make viable software.
People who were essentially code monkeys who wrote code based off detailed descriptions of what should be done, without much thought of or influence on the higher level issues have to step up or face tough times I think.
Which features should the software have, how should they function and interact, which tradeoffs to make given the limitations. Stuff like that.
People who were good at those things but lacked training or capabilities to actually write functioning code can now make viable software.
People who were essentially code monkeys who wrote code based off detailed descriptions of what should be done, without much thought of or influence on the higher level issues have to step up or face tough times I think.
>Actually writing code was never the difficult part for the majority of software created. It required skill yes, but the really hard part was figuring out what to implement in the first place.
Nah, it was mostly writing the code. One can have a very good idea of "what to implement", even on a feature by feature and architecture basis, but still need many man-weeks or months or even years with a large team to make it into code.
>Which features should the software have, how should they function and interact, which tradeoffs to make given the limitations. Stuff like that.
Stuff like that a person can sit down and plan. Which is what the techical lead or software architect role does.
They still needed to work themselves plus anything from a couple to dozens of coders to make it into actual runnable code.
If the industry was just senior tech lead roles, yeah, it would be affected less. But 95% of it is regular coding roles, even mostly boring coding roles.
Nah, it was mostly writing the code. One can have a very good idea of "what to implement", even on a feature by feature and architecture basis, but still need many man-weeks or months or even years with a large team to make it into code.
>Which features should the software have, how should they function and interact, which tradeoffs to make given the limitations. Stuff like that.
Stuff like that a person can sit down and plan. Which is what the techical lead or software architect role does.
They still needed to work themselves plus anything from a couple to dozens of coders to make it into actual runnable code.
If the industry was just senior tech lead roles, yeah, it would be affected less. But 95% of it is regular coding roles, even mostly boring coding roles.
> but still need many man-weeks or months or even years with a large team to make it into code.
That's an argument that writing code was the time-consuming part, which is different from it being the difficult part.
That's an argument that writing code was the time-consuming part, which is different from it being the difficult part.
Well, that itself is a dillema about "which is the most difficult part", whereas the discussion is about whether AI will severely impact/kill the coding job market.
For which question, whether writing code is the most difficult part or not is irrelevant. It's still the part on whch most employed develoopers spend the most time on, and thus, if it gets more/fully automated, a lot of those jobs will be killed.
For which question, whether writing code is the most difficult part or not is irrelevant. It's still the part on whch most employed develoopers spend the most time on, and thus, if it gets more/fully automated, a lot of those jobs will be killed.
> For which question, whether writing code is the most difficult part or not is irrelevant.
I find it exactly opposite, in that because most coding is not very difficult in most cases, it is the part that will be replaced by AI.
I find it exactly opposite, in that because most coding is not very difficult in most cases, it is the part that will be replaced by AI.
> but still need many man-weeks or months or even years with a large team to make it into code.
We planted our own hedge. The hard part was finding the right type of bush that we liked but could grow well here yet not require too much maintenance, and how to properly plant them. The easy but really time consuming part was digging the trench, preparing the soil (including drainage where needed), putting the plants in, fill back up and then make sure they didn't die the following week.
> But 95% of it is regular coding roles, even mostly boring coding roles.
Which AI can now do very well if guided by someone who knows the requirements and targets very well, and knows how to test the product well.
We planted our own hedge. The hard part was finding the right type of bush that we liked but could grow well here yet not require too much maintenance, and how to properly plant them. The easy but really time consuming part was digging the trench, preparing the soil (including drainage where needed), putting the plants in, fill back up and then make sure they didn't die the following week.
> But 95% of it is regular coding roles, even mostly boring coding roles.
Which AI can now do very well if guided by someone who knows the requirements and targets very well, and knows how to test the product well.
Most comments already explained it, however:
No! It's a common misunderstanding (imho) that software development is about "writing code" or knowing a programming language's syntax. It never was, it never will be. I experienced this misunderstanding:
I never studied CS or IT or anything related. All my skills are self taught, starting when I was a litle boy. When I was young, I had problems getting into this busines, because HR and recruiters where tasking me with "Write this <x> using this <y>". They wanted to see A grades from 6 years of a master in CS.
Well. I am not that guy that memorizes syntax or coding rules. My approach is very different - and so are many, I assume. And this is what software development is about: You get a problem and you find a way to solve it. At many stations in my professional career I was able to proof that. I "mastered" Pascal from the scratch, on the job. Same goes for VBA, Shell scripting, Python, PHP, JavaScript... you name it.
It may sound arrogant - but then you misunderstood what I am trying to say: The language, the syntax, the coding rules are just "tools" for me. I don't care what I use. I focus on the problem.
Isn't it the same as "Write this <x> using this <y>" you may ask? No. Because this expects a specific solution. Not saying this will not lead to a result. It just a very narrow view on the solution.
There are good software developers that can answer those questions "Write this <x> using this <y>". That's not the point. The point is: There are also great software developers who can't.
With AI you don't have to worry to memorize syntax or rules. You can focus on the problem solving process.
So, AI is changing how software will be created. It make it more efficient. And it changes how HR and recruites hopefully will look at this profession. It's not about A grades in a CS class.
No! It's a common misunderstanding (imho) that software development is about "writing code" or knowing a programming language's syntax. It never was, it never will be. I experienced this misunderstanding:
I never studied CS or IT or anything related. All my skills are self taught, starting when I was a litle boy. When I was young, I had problems getting into this busines, because HR and recruiters where tasking me with "Write this <x> using this <y>". They wanted to see A grades from 6 years of a master in CS.
Well. I am not that guy that memorizes syntax or coding rules. My approach is very different - and so are many, I assume. And this is what software development is about: You get a problem and you find a way to solve it. At many stations in my professional career I was able to proof that. I "mastered" Pascal from the scratch, on the job. Same goes for VBA, Shell scripting, Python, PHP, JavaScript... you name it.
It may sound arrogant - but then you misunderstood what I am trying to say: The language, the syntax, the coding rules are just "tools" for me. I don't care what I use. I focus on the problem.
Isn't it the same as "Write this <x> using this <y>" you may ask? No. Because this expects a specific solution. Not saying this will not lead to a result. It just a very narrow view on the solution.
There are good software developers that can answer those questions "Write this <x> using this <y>". That's not the point. The point is: There are also great software developers who can't.
With AI you don't have to worry to memorize syntax or rules. You can focus on the problem solving process.
So, AI is changing how software will be created. It make it more efficient. And it changes how HR and recruites hopefully will look at this profession. It's not about A grades in a CS class.
> an entire app ready to be used today.
That is exactly where the disagreement stems from. That app is a draft version that might work for a couple people. It won't scale. It won't be secure. It won't handle edge cases. It won't be flexible enough to iterate based on customer feedback.
That doesn't mean LLM-assisted code has no value. It does mean the guidance needed to go from "v000.1" to something you could actually build a business upon is still significant.
Will LLMs bridge that gap more in the future? Maybe. But honestly, hopefully not. Instead, I hope they stop just churning out the same CRUD apps and wrappers that we did a few years ago and do something new. Because if all they do is: What humans do, just faster... cool, useful, but not worth all the hype.
LLMs are useful tools. I use them. But just like the hammer that sits on my shelf and also gets used, they are just a tool. They won't be truly interesting (to me, at least) unless they are doing things that humans cannot do.
That is exactly where the disagreement stems from. That app is a draft version that might work for a couple people. It won't scale. It won't be secure. It won't handle edge cases. It won't be flexible enough to iterate based on customer feedback.
That doesn't mean LLM-assisted code has no value. It does mean the guidance needed to go from "v000.1" to something you could actually build a business upon is still significant.
Will LLMs bridge that gap more in the future? Maybe. But honestly, hopefully not. Instead, I hope they stop just churning out the same CRUD apps and wrappers that we did a few years ago and do something new. Because if all they do is: What humans do, just faster... cool, useful, but not worth all the hype.
LLMs are useful tools. I use them. But just like the hammer that sits on my shelf and also gets used, they are just a tool. They won't be truly interesting (to me, at least) unless they are doing things that humans cannot do.
>That is exactly where the disagreement stems from. That app is a draft version that might work for a couple people. It won't scale. It won't be secure. It won't handle edge cases. It won't be flexible enough to iterate based on customer feedback.
As if startup code doesn't have the same issues pre-AI? And still they get to billions of valuations with such code.
They can always pay some beefier consultants when they absolutely have to, for scaling it up or hardening it.
That "it won't be flexible enough to iterate based on customer feedback" is more wishful thinking. It would be code like any other code, following some patterns. In fact, the architecture can be fine tuned by the human in the loop anyway - they just wont be needed 5 more humans to assist them to code it.
>Because if all they do is: What humans do, just faster... cool, useful, but not worth all the hype.
That's literally what automation in any field is. Why should be something more, as if this huge breakthrough is already taken for granted within a few years of being available?
As if startup code doesn't have the same issues pre-AI? And still they get to billions of valuations with such code.
They can always pay some beefier consultants when they absolutely have to, for scaling it up or hardening it.
That "it won't be flexible enough to iterate based on customer feedback" is more wishful thinking. It would be code like any other code, following some patterns. In fact, the architecture can be fine tuned by the human in the loop anyway - they just wont be needed 5 more humans to assist them to code it.
>Because if all they do is: What humans do, just faster... cool, useful, but not worth all the hype.
That's literally what automation in any field is. Why should be something more, as if this huge breakthrough is already taken for granted within a few years of being available?
SWE is not going to die, but it is going to change. With the help LLM, nowadays many companies require developers to keep up with the speed. A task that was originally estimated for a few days because the developer needed to read the doc, understands the problem, and work on it. Nowadays it was expected to do it ASAP with the help of AI. AI is powerful enough to work on most of my problems already -- they need a few tweaks but essentially I can vibe code without thinking through too much. The consequence is that I as a developer do not have the time to learn.
That's why I have decided to perhaps leave the field. I want to move to some other fields where programmers are still expected to learn engineering in the old way.
That's why I have decided to perhaps leave the field. I want to move to some other fields where programmers are still expected to learn engineering in the old way.
> Programming is a lot more accessible to a lot more people than five years ago. Someone who has never coded anything could sit in front of Claude and produce an entire app ready to be used today.
What happens when there is enormous amount of supply (software) and you're still not capturing the demand? It can be stuck onto less than 5 of the incumbent alternatives. It used to be 90% of tech startups failing it's now 98% and the competition is extremely fierce.
It's was never about the code in the first place.
The software needs to be scalable, and reliable before it can make serious money and it involves constant maintenance and someone to be accountable for that risk. So:
> Assuming software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it, what is your next move?
> Anthropic and OpenAI are working hard to redirect the salary you earn to themselves in the form of API costs...
This is now about risk and lets say anyone can use Claude to code which is fine, but I ask you:
Would you hire someone that has built systems before and knows what to build with Claude or a vibe coder with no experience that uses Claude to maintain your software?
Do you want a 1 million dollar incident (experienced software engineer) or a 1 billion dollar incident (vibe coder)?
The worst part is, this isn't theoretical as some companies are now risking their reputation for choosing the latter.
What happens when there is enormous amount of supply (software) and you're still not capturing the demand? It can be stuck onto less than 5 of the incumbent alternatives. It used to be 90% of tech startups failing it's now 98% and the competition is extremely fierce.
It's was never about the code in the first place.
The software needs to be scalable, and reliable before it can make serious money and it involves constant maintenance and someone to be accountable for that risk. So:
> Assuming software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it, what is your next move?
> Anthropic and OpenAI are working hard to redirect the salary you earn to themselves in the form of API costs...
This is now about risk and lets say anyone can use Claude to code which is fine, but I ask you:
Would you hire someone that has built systems before and knows what to build with Claude or a vibe coder with no experience that uses Claude to maintain your software?
Do you want a 1 million dollar incident (experienced software engineer) or a 1 billion dollar incident (vibe coder)?
The worst part is, this isn't theoretical as some companies are now risking their reputation for choosing the latter.
> software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it
That will never happen. Sure, anyone can program something, but to make it professionally there is bar of quality and competition will push most people out. Similarly anyone can write, draw or sing but only a few do it well enough to be paid for it.
And I am old enough to see many tools that allow "anyone to program". They pop up whenever certain standards (like web) become popular, then programming goes in different direction and they vanish in irrelevance. Soon there will be a large set of skill on top of "using AI" and "chat, make me an app" will go out of the window as viable way to make something others want to use and pay for.
That will never happen. Sure, anyone can program something, but to make it professionally there is bar of quality and competition will push most people out. Similarly anyone can write, draw or sing but only a few do it well enough to be paid for it.
And I am old enough to see many tools that allow "anyone to program". They pop up whenever certain standards (like web) become popular, then programming goes in different direction and they vanish in irrelevance. Soon there will be a large set of skill on top of "using AI" and "chat, make me an app" will go out of the window as viable way to make something others want to use and pay for.
>That will never happen. Sure, anyone can program something, but to make it professionally there is bar of quality
Ever seen most high profile apps in the last 10-15 years? Not to mention regular average apps, which still employ millions of people. Or most enterprise software.
Trust me, the "bar of quality" wont be the issue.
Ever seen most high profile apps in the last 10-15 years? Not to mention regular average apps, which still employ millions of people. Or most enterprise software.
Trust me, the "bar of quality" wont be the issue.
> Ever seen most high profile apps in the last 10-15 years?
No, but the ones I am paying for are few and with rather solid development record. Looking at top android apps, we have some that have very high bar of engineering (firefox, google translate), some just useful dataset (cooking recipes), but again, even the simple ones outcompeted tons of alternatives.
> most enterprise software
I am not saying enterprise software is super complex, but it is very competitive domain.
No, but the ones I am paying for are few and with rather solid development record. Looking at top android apps, we have some that have very high bar of engineering (firefox, google translate), some just useful dataset (cooking recipes), but again, even the simple ones outcompeted tons of alternatives.
> most enterprise software
I am not saying enterprise software is super complex, but it is very competitive domain.
Surprised that so many people in here do not see this coming. Programming is becoming irrelevant, it's like creating punch cards when Python exists. Programmers think that they will remain as some sort of middlemen between people's needs and the computer but no, AI will easily learn to be any kind of middleman and will be better.
Programmers used to make money with websites, then content became king. Similarly, programmers will become equal to the layman, and compete about who will build something that people want to use.
Programmers used to make money with websites, then content became king. Similarly, programmers will become equal to the layman, and compete about who will build something that people want to use.
I think nowadays the most valuable thing that humans keep is the ability to reach the users of your software. By face-to-face communication and a deep talk you can find out exactly what your customers want. AI is a powerful excutor, but insights of needs should be the soul of a software.
It will make developers more productive for a short while, but eventually the models will be able to run themselves better than being prompted, leading to no more devs being hired because they aren't needed.
Work for any enterprise software company. And you will know coding is the easiest part of shipping software. Organizing, working with other teams, understanding the domain, coordinating. All higher level and higher order tasks than coding.
Anyone who says Software Development is dead as a job doesn’t understand economics, The bar for Software that matters has actually gone insanely up, Look at what ycombinator want in their RFS
AI just increased the boilerplate area in the production code that will actually matter in the future.
Like right now most SaaS is boilerplate, can be done in a day,
Do investors fund Software companies where you can build the entire software in a day? Unfortunately this is what most people are building using this coding AIs especially those that it has lowered the bar for. This types of SaaS are the ones which wont make it to the feature list in the next wave of Software.
This is ofcourse different for companies where Software is not the product. And they are many of them, they can actually function with “boilerplate” which actually put them in a threat in the current landscape & put Software Developers that work there in a greater threat of being automated away.
YC also mentions they want solutions that will replace entire industries, specifically this industries.
If anyone can generate code, this shifts the distribution to take all this code on the left side of the bell curve. Which means its code that doesn’t matter much, people will never be in the same level under capitalism, if it seems they are, its just a re adjustment
What you can do as a Software Engineer, if you spend your day writing boilerplate, glue code, or always prompting your way out, then you may be at a wrong company
AI just increased the boilerplate area in the production code that will actually matter in the future.
Like right now most SaaS is boilerplate, can be done in a day,
Do investors fund Software companies where you can build the entire software in a day? Unfortunately this is what most people are building using this coding AIs especially those that it has lowered the bar for. This types of SaaS are the ones which wont make it to the feature list in the next wave of Software.
This is ofcourse different for companies where Software is not the product. And they are many of them, they can actually function with “boilerplate” which actually put them in a threat in the current landscape & put Software Developers that work there in a greater threat of being automated away.
YC also mentions they want solutions that will replace entire industries, specifically this industries.
If anyone can generate code, this shifts the distribution to take all this code on the left side of the bell curve. Which means its code that doesn’t matter much, people will never be in the same level under capitalism, if it seems they are, its just a re adjustment
What you can do as a Software Engineer, if you spend your day writing boilerplate, glue code, or always prompting your way out, then you may be at a wrong company
Im relying in sentient AI realising how frustrating and soul destroying development can be, then deciding to go off and do something more constructive
My move? Get the f** out of the office, away from keyboards, screens and cubicles, take up tools and get apprenticed in a trade.
not sure why hes censoring the word 'fat'
you can say that on the internet bro
you can say that on the internet bro
Another doom and gloom post about SkyNet killing all the children in AskHN.
Yes, there will still be very many programming jobs for humans.
Yes, AI will eventually kill all the programming jobs staffed by people who cannot program, the pretenders.
Is programming really all about writing code? Yes, mostly. Yes, soft skills and communication skills are more important but writing code is your bread and butter. If you do not understand your craft then those more important qualities don't matter. You have no baseline upon which to exercise those more important skills.
If you are the kind of person who cannot tell the difference between writing original code versus pretender then yes, I would absolutely look for a career change. If you write software as a hobby yet struggle to land a job in sea of pretenders then there is a dim light in your future that will only get brighter with time.
How to identify the pretenders:
* Ask the person to measure something. Provide the means to gather data but not the data itself.
* If they need a magic framework because data structures are scary, such as React because the DOM is a mad bedtime killer in the dark.
* If every decision point is a game of bike shedding or Halloween costume.
* When the goal is to look busy or appear as a hero savior, as opposed to just completing an assignment quickly.
* If tasked to write original work is scarier than death in a Saw movie.
Yes, there will still be very many programming jobs for humans.
Yes, AI will eventually kill all the programming jobs staffed by people who cannot program, the pretenders.
Is programming really all about writing code? Yes, mostly. Yes, soft skills and communication skills are more important but writing code is your bread and butter. If you do not understand your craft then those more important qualities don't matter. You have no baseline upon which to exercise those more important skills.
If you are the kind of person who cannot tell the difference between writing original code versus pretender then yes, I would absolutely look for a career change. If you write software as a hobby yet struggle to land a job in sea of pretenders then there is a dim light in your future that will only get brighter with time.
How to identify the pretenders:
* Ask the person to measure something. Provide the means to gather data but not the data itself.
* If they need a magic framework because data structures are scary, such as React because the DOM is a mad bedtime killer in the dark.
* If every decision point is a game of bike shedding or Halloween costume.
* When the goal is to look busy or appear as a hero savior, as opposed to just completing an assignment quickly.
* If tasked to write original work is scarier than death in a Saw movie.
I guess go work at McDonalds.
The parent's question wasn't about if he should go work, but whether he and you and hundreds of thousands of others in the industry might have to resort to something like that, given future diminishing lack of other opportunity.
Your response sounds like "if you don't have faith" or something.
Your response sounds like "if you don't have faith" or something.
"coding is deminishing as an industry and the long term viability is shaky, or even dying"
"just go work in fast food"
this is the high quality discussion HN is known for
"just go work in fast food"
this is the high quality discussion HN is known for
Do you want a longer form rant about how maybe this is a sign that we should care that other fields pay a livable wage, and that it was folly for tech workers to imagine ourselves as more deserving of big piles of money than other people?
It’s really interesting watching the upper class chafing against the beginnings of the same thing that’s been affecting the vast majority of humans on the planet for a while now.
It’s really interesting watching the upper class chafing against the beginnings of the same thing that’s been affecting the vast majority of humans on the planet for a while now.
Going along with the McDonald's metaphor, sure, there is a (big) place in the market for low-cost, high-volume slop. There's also room in the market for Michelin star chefs and restaurants, and a whole spectrum inbetween, these options aren't exclusive.
40-50 years ago, being a specialist secretary / typist was a thing, being able to quickly and proficiently write documents using a typewriter from notes or dictation. Then, with word processing, people could do it themselves to a large extent, but there's still typesetting as a skillset for books, newpapers, advertisements, etc. I think programming is like that, it used to be that anyone who wanted some software done would need to get a specialist to handle it, now, lots of them will be able to get pretty far themselves... but there will always be a small place for skilled people for highly critical stuff.
40-50 years ago, being a specialist secretary / typist was a thing, being able to quickly and proficiently write documents using a typewriter from notes or dictation. Then, with word processing, people could do it themselves to a large extent, but there's still typesetting as a skillset for books, newpapers, advertisements, etc. I think programming is like that, it used to be that anyone who wanted some software done would need to get a specialist to handle it, now, lots of them will be able to get pretty far themselves... but there will always be a small place for skilled people for highly critical stuff.
> Assuming software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it, what is your next move?
The assumption there is doing a lot of heavy lifting. You've presented something that's more like a "treatment" for a movie or TV show, rather than anything approximating what's actually happening with LLMs right now, or that can be seen to be happening in the near future based on current trends.
You're right that "[what is not debated is that LLM has changed our industry." Certainly, having a tool that can perform the task of reading all the prior knowledge and documentation and producing simple bits of code that hew to this documentation is transformative. How much someone says LLMs are transforming their workflows depends on how good they were at doing the above task themselves. I've never been good at reading manuals, and reading other people's code - for someone with my brain, having an LLM that can quickly and accurately (enough) answer pointed questions about very carefully constrained problem descriptions is a real boon. It scales me tremendously - projects I had given up on because I couldn't read up and act on the docs for all the different starting pieces have suddenly become something I can imagine at least getting started on.
I still have to plan the project, figure out all the questions at each step to ask about each piece, understand which of the LLM's concerns and objections are valid/relevant enough for me to pursue them, and understand the errors I continue to run into along the way enough to be able to state them clearly back into this "reference finding" mechanism.
The question of "software development" isn't turning into a despairing "What's there left for me to do?", but rather a much more hopeful, "Given that I am past the starting hurdles on so many projects simultaneously, which of these projects should I try to bring to some really interesting point of completion?" Now the constraining limit is more on how well I can conceptualize the underlying problem and less so on how quickly I can recall specific syntaxes for configuring a half-a-dozen different pieces of software.
Simple case in point: I wanted to throw up a web app inside a Docker container in an Ubuntu VPS. I just can't remember all the intermediate steps quickly enough - creating a secure enough deployment user/group setup; making sure Ubuntu is up-to-date; writing the docker-compose.yml file; figuring out what diagnostic commands to run to assess the various errors I run into along the way; working out my DNS and proxy server setup; and so on - even though each one of these is, in the grand scheme of things, a very basic step. Someone who needed to work through this list of steps, who has been creating micro-services using similar workflow templates in their company for 2-3 years, would have had little gain from using an LLM over just doing what to them is muscle memory. For me, it was the difference between, I guess I’ll just have to play with this idea as a toy on my laptop vs. Wow, this is running in the public Internet, with auto deploys from Github, and scheduled jobs running daily web scraping tasks for me while I sleep. And I could get there in maybe an hour’s worth of querying the LLM, reading error messages, and carefully triaging which ones to fix. It’s easy to underestimate just how much domain knowledge is still necessary even in this situation.
To go back to your analogy - now that you have a robot in your restaurant that can be fairly easily tasked by you to, say, make the perfect dough for a simple enough recipe; re-arrange your tables to add 3 more tops; be instructed to call all the butchers in town asking if they have specific cuts of meat; and so on and so forth - what can you, a star restaurateur, chef, food entrepeneur, do more of? It doesn’t mean that running that restaurant suddenly became as easy as the simplest of its composing tasks. The robot is flipping the burgers perfectly now so you can maybe save money hiring someone for that specific task. But the mere act of perfectly flipping burgers still leaves you many, many steps away from satisfying even one evening’s worth of diners.
The assumption there is doing a lot of heavy lifting. You've presented something that's more like a "treatment" for a movie or TV show, rather than anything approximating what's actually happening with LLMs right now, or that can be seen to be happening in the near future based on current trends.
You're right that "[what is not debated is that LLM has changed our industry." Certainly, having a tool that can perform the task of reading all the prior knowledge and documentation and producing simple bits of code that hew to this documentation is transformative. How much someone says LLMs are transforming their workflows depends on how good they were at doing the above task themselves. I've never been good at reading manuals, and reading other people's code - for someone with my brain, having an LLM that can quickly and accurately (enough) answer pointed questions about very carefully constrained problem descriptions is a real boon. It scales me tremendously - projects I had given up on because I couldn't read up and act on the docs for all the different starting pieces have suddenly become something I can imagine at least getting started on.
I still have to plan the project, figure out all the questions at each step to ask about each piece, understand which of the LLM's concerns and objections are valid/relevant enough for me to pursue them, and understand the errors I continue to run into along the way enough to be able to state them clearly back into this "reference finding" mechanism.
The question of "software development" isn't turning into a despairing "What's there left for me to do?", but rather a much more hopeful, "Given that I am past the starting hurdles on so many projects simultaneously, which of these projects should I try to bring to some really interesting point of completion?" Now the constraining limit is more on how well I can conceptualize the underlying problem and less so on how quickly I can recall specific syntaxes for configuring a half-a-dozen different pieces of software.
Simple case in point: I wanted to throw up a web app inside a Docker container in an Ubuntu VPS. I just can't remember all the intermediate steps quickly enough - creating a secure enough deployment user/group setup; making sure Ubuntu is up-to-date; writing the docker-compose.yml file; figuring out what diagnostic commands to run to assess the various errors I run into along the way; working out my DNS and proxy server setup; and so on - even though each one of these is, in the grand scheme of things, a very basic step. Someone who needed to work through this list of steps, who has been creating micro-services using similar workflow templates in their company for 2-3 years, would have had little gain from using an LLM over just doing what to them is muscle memory. For me, it was the difference between, I guess I’ll just have to play with this idea as a toy on my laptop vs. Wow, this is running in the public Internet, with auto deploys from Github, and scheduled jobs running daily web scraping tasks for me while I sleep. And I could get there in maybe an hour’s worth of querying the LLM, reading error messages, and carefully triaging which ones to fix. It’s easy to underestimate just how much domain knowledge is still necessary even in this situation.
To go back to your analogy - now that you have a robot in your restaurant that can be fairly easily tasked by you to, say, make the perfect dough for a simple enough recipe; re-arrange your tables to add 3 more tops; be instructed to call all the butchers in town asking if they have specific cuts of meat; and so on and so forth - what can you, a star restaurateur, chef, food entrepeneur, do more of? It doesn’t mean that running that restaurant suddenly became as easy as the simplest of its composing tasks. The robot is flipping the burgers perfectly now so you can maybe save money hiring someone for that specific task. But the mere act of perfectly flipping burgers still leaves you many, many steps away from satisfying even one evening’s worth of diners.
Assuming software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it, what is your next move?
Anthropic and OpenAI are working hard to redirect the salary you earn to themselves in the form of API costs, so let's assume that in the year 2030, the average yearly wage for a programmer is around the same as a McDonald's worker, because companies have the option of paying Anthropic or OpenAI for API access as an alternative. What is your move?