> One of the things I’ve come to appreciate about coding with LLMs is I have greater control over how I want the code to work and be designed.
I don't get it. Writing code yourself is the best way to control how the code functions and is designed. If LLMs were to disappear tomorrow, how would one lose control over the degree and specificity of the code produced that one could not simply compensate for with a bit more time and skill investment up front?
The author only hints at the notion that developers who do not use LLMs are like painters upon the advent of photography, but does not go on to suggest who these developers analogous to painters might be today, nor to whom their works might be valued.
If we're going to torture visual media into being a good analogy for LLM versus hand coding, I think writing code oneself is more like having a inkjet printer that one controls manually. You get a great deal of explicit control and theoretically unlimited expressiveness but at the cost of a greater time investment. Comparatively, painting might not give you that same degree of precision. Photography surrenders that control but puts a great significance on capturing fleeting moments in time.
The article's argument is pretty common - but it's looking for a precedent for a technological advance without a precedent. Cameras weren't capable of reasoning or thinking (or for the quibblers, predicting output well enough to be indiscernible from reasoning and thinking), and they weren't generalized.
I'm not a senior engineer and I trust those that tell me LLMs are currently incapable of high level coherent engineering. But at the same time, this is a rapidly developing tech whose progress only seems to be accelerating, and as long as it's gaining on every area of coding and planning, anyone arguing AI will never be equal or superior to human SWEs is wish thinking.
We've lived through all this before with the Go community, who used to be adamant a machine could never do better than a human - because of some ethereal semi-magical quality only existing in the skulls of bipeds and totally inaccessible to machines. SWE's Alphago Zero moment could be in the post.
When is the last time, other than one of those coding on a whiteboard interviews, where you coded your own implementation of an important algorithm?
AI has created more situations where an algorithmic twist or a really clever implementation makes a significant difference. But those situations are still rare compared to everyday work in software development.
The real danger is AI slop displacing actual good resources for software engineering. Once those are gone they will have to be reverse engineered from the AI it's weights.
I'd argue the kinds of things that LLMs spit out these days don't actually require "software engineering" and haven't for a while.
If you can slop it out with a coding agent it's probably not the kind of thing that needed an engineer in the first place, just someone who can write the code. Which is not the same thing.
You certainly can do software engineering with LLMs (and in some cases it supercharges it). But then you're doing actual software engineering anyway, with or without the LLM.
The only thing that's changed is being able to call yourself a software engineer when you weren't any such thing in the first place. I agree it's harder to do that now.
> Before LLMs I would have to spend weeks or more implementing a design decision, and maybe weeks or months if I had to refactor one
I wonder why people think that is a good argument. Spending weeks on a single design decision makes the case that you’re either not a good developer, your time was stolen by other concerns, or that design decision was a major process change. In the last case, this is one of the goals of software engineering to make those less painful.
Anyone using this argunent should add anecdotes so that others can appreciate the previous scope of work.
> One of the things I’ve come to appreciate about coding with LLMs is I have greater control over how I want the code to work and be designed.
I don't get it. Writing code yourself is the best way to control how the code functions and is designed. If LLMs were to disappear tomorrow, how would one lose control over the degree and specificity of the code produced that one could not simply compensate for with a bit more time and skill investment up front?
The author only hints at the notion that developers who do not use LLMs are like painters upon the advent of photography, but does not go on to suggest who these developers analogous to painters might be today, nor to whom their works might be valued.
If we're going to torture visual media into being a good analogy for LLM versus hand coding, I think writing code oneself is more like having a inkjet printer that one controls manually. You get a great deal of explicit control and theoretically unlimited expressiveness but at the cost of a greater time investment. Comparatively, painting might not give you that same degree of precision. Photography surrenders that control but puts a great significance on capturing fleeting moments in time.
The article's argument is pretty common - but it's looking for a precedent for a technological advance without a precedent. Cameras weren't capable of reasoning or thinking (or for the quibblers, predicting output well enough to be indiscernible from reasoning and thinking), and they weren't generalized.
I'm not a senior engineer and I trust those that tell me LLMs are currently incapable of high level coherent engineering. But at the same time, this is a rapidly developing tech whose progress only seems to be accelerating, and as long as it's gaining on every area of coding and planning, anyone arguing AI will never be equal or superior to human SWEs is wish thinking.
We've lived through all this before with the Go community, who used to be adamant a machine could never do better than a human - because of some ethereal semi-magical quality only existing in the skulls of bipeds and totally inaccessible to machines. SWE's Alphago Zero moment could be in the post.
When is the last time, other than one of those coding on a whiteboard interviews, where you coded your own implementation of an important algorithm?
AI has created more situations where an algorithmic twist or a really clever implementation makes a significant difference. But those situations are still rare compared to everyday work in software development.
Long live Software Engineering!
The real danger is AI slop displacing actual good resources for software engineering. Once those are gone they will have to be reverse engineered from the AI it's weights.
I'd argue the kinds of things that LLMs spit out these days don't actually require "software engineering" and haven't for a while.
If you can slop it out with a coding agent it's probably not the kind of thing that needed an engineer in the first place, just someone who can write the code. Which is not the same thing.
You certainly can do software engineering with LLMs (and in some cases it supercharges it). But then you're doing actual software engineering anyway, with or without the LLM.
The only thing that's changed is being able to call yourself a software engineer when you weren't any such thing in the first place. I agree it's harder to do that now.
We moved from developers to janitors, I fear. No disrespect to janitors who do great work whereas babysitting AI slop generation isn't respectful.
> Before LLMs I would have to spend weeks or more implementing a design decision, and maybe weeks or months if I had to refactor one
I wonder why people think that is a good argument. Spending weeks on a single design decision makes the case that you’re either not a good developer, your time was stolen by other concerns, or that design decision was a major process change. In the last case, this is one of the goals of software engineering to make those less painful.
Anyone using this argunent should add anecdotes so that others can appreciate the previous scope of work.
Why do you think what you said was a good argument. You said, "hes dumb or distracted or deserves it" You just made lazy moralizing arguments.