It wasn't apparent to me before the advent of LLMs that many of my colleagues think very differently about software. Obviously, the primary goal of software is to Do Thing, but to me, a lot of the satisfaction of my job comes from kneading the code until it's as streamlined, simple and elegant as I can make it while accumulating a thorough understanding of the entire system.
I don't like how coding agents disconnect me from that process.
For me, a lot of what I enjoy is coding is the process between coming up with a plan, and actually seeing the final result. I internalize and understand code much, much better when I write it.
When my job is primarily reading code, and tidying up after, the joy of incremental improvement is gone, and I'm left with something that doesn't feel like mine. I don't feel like I built any part of AI generated code, even if I guided every part. It turns me into a manager, rather than a contributor, and that's not where I enjoy being.
Though the rise of AI has given me a bit of a reality check in that most people and companies don't care about how the sausage gets made, only that it does, and that it's done with the utmost efficiency, even if it results in something that's lower quality. It just has to be good enough.
I was worried I wouldn't understand the output—that I'd lose my ability to judge whether the code was actually right. But it turns out decades of reading and reviewing code doesn't evaporate.
What about the next generation? How will they get their decades of this experience?
This is a fair response for a certain frame of thinking about the world. It's certainly optimistic, and I am thankful that there are optimistic people around.
When calculators came along, the ability to perform written & mental maths was eroded, but didn't vanish completely. We'll see the same with AI and raw coding skills.
Same with assembly and "current generation" (whatever it would mean). Few write and understand it, but it exists out there. People still learn it when necessary.
I find it to be an unfair comparison because calculator are not able to do math on their own if you ask them a question and they are not able to search for formulas or theorem, etc...
For example at work, my product manager can create a working PoC of an app she has thought of using an LLM while having 0 knownledge about how it's working. But she would'nt be able to achieve anything with a calculator (or even a computer) if she thought about some statistical problem to solve.
My point is that I find that something is very off when say the rise of LLM and how we will adapt to it is similar to the rise of C vs assembly or the rise of calculator vs mental calculus. In the former you can have outcomes without the need to acquire any knowledge.
I think an LLM today would make me pretty much succeed all the engineering exams I had to do at school but without having to actually learn anything. There is a motive to learn something in order to achieve it, but if the thing can be done without learning what motive to learn is there left ?
Your product manager has enormous depth of experience and knowledge about apps that she can use as part of building that proof of concept - that's the nature of being a PM.
She may have been missing the low level coding knowledge but it's important not to discount all of that high level product design knowledge.
I've seen this analogy before. It's not a bad one, in some sense, but I don't agree with it. There are too many qualities that make this a completely different scenario, and not all technologies are the same.
fleebee | a day ago
These are good observations about "the split".
It wasn't apparent to me before the advent of LLMs that many of my colleagues think very differently about software. Obviously, the primary goal of software is to Do Thing, but to me, a lot of the satisfaction of my job comes from kneading the code until it's as streamlined, simple and elegant as I can make it while accumulating a thorough understanding of the entire system.
I don't like how coding agents disconnect me from that process.
chai | 22 hours ago
For me, a lot of what I enjoy is coding is the process between coming up with a plan, and actually seeing the final result. I internalize and understand code much, much better when I write it.
When my job is primarily reading code, and tidying up after, the joy of incremental improvement is gone, and I'm left with something that doesn't feel like mine. I don't feel like I built any part of AI generated code, even if I guided every part. It turns me into a manager, rather than a contributor, and that's not where I enjoy being.
Though the rise of AI has given me a bit of a reality check in that most people and companies don't care about how the sausage gets made, only that it does, and that it's done with the utmost efficiency, even if it results in something that's lower quality. It just has to be good enough.
darth-cheney | 22 hours ago
What about the next generation? How will they get their decades of this experience?
simonw | 22 hours ago
I expect they'll figure that out themselves.
Being able to produce high quality, maintainable, bug-free software is a valuable skill to possess that will be rewarded by the job market.
If you're a new engineer you are competing with your peers to develop those skills. This means you're incentivized to figure out how to best learn.
I feel like a lot of the concern people have for the next generation is failing to acknowledge their own agency in this matter.
I'm ready to help that generation in any way I can, but I have a suspicion that they might not need as much help as people assume they will.
darth-cheney | 21 hours ago
This is a fair response for a certain frame of thinking about the world. It's certainly optimistic, and I am thankful that there are optimistic people around.
alexjurkiewicz | 17 hours ago
When calculators came along, the ability to perform written & mental maths was eroded, but didn't vanish completely. We'll see the same with AI and raw coding skills.
viraptor | 14 hours ago
Same with assembly and "current generation" (whatever it would mean). Few write and understand it, but it exists out there. People still learn it when necessary.
kevinc | 13 hours ago
"The Russians used a keyboard"
(Though I'm conflating the pocket calculator and the space pen.)
gerikson | 10 hours ago
The fun anecdote about the thrifty Soviets vs the profligate Americans is unfortunately(?) a myth:
https://www.scientificamerican.com/article/fact-or-fiction-nasa-spen/
ducdetronquito | 10 hours ago
I find it to be an unfair comparison because calculator are not able to do math on their own if you ask them a question and they are not able to search for formulas or theorem, etc...
For example at work, my product manager can create a working PoC of an app she has thought of using an LLM while having 0 knownledge about how it's working. But she would'nt be able to achieve anything with a calculator (or even a computer) if she thought about some statistical problem to solve.
My point is that I find that something is very off when say the rise of LLM and how we will adapt to it is similar to the rise of C vs assembly or the rise of calculator vs mental calculus. In the former you can have outcomes without the need to acquire any knowledge.
I think an LLM today would make me pretty much succeed all the engineering exams I had to do at school but without having to actually learn anything. There is a motive to learn something in order to achieve it, but if the thing can be done without learning what motive to learn is there left ?
simonw | 9 hours ago
Your product manager has enormous depth of experience and knowledge about apps that she can use as part of building that proof of concept - that's the nature of being a PM.
She may have been missing the low level coding knowledge but it's important not to discount all of that high level product design knowledge.
darth-cheney | 7 hours ago
I've seen this analogy before. It's not a bad one, in some sense, but I don't agree with it. There are too many qualities that make this a completely different scenario, and not all technologies are the same.
carlana | a day ago
Deep blue…
marginalia | a day ago
sigmonsez | 6 hours ago
What about a few giant corporations being inserted into the development process who will report to a board and continue to raise prices?
Having a few capitalistic companies blocking software development creates a very grim future.