Thanks for reading / commenting this post. Initially it seemed like I received a bunch of very negative comments, now I read most of the thread, and there are very good points, articulated with sensibility. Thank you.
I wanted to provide some more context that is not part of the blog post. Since somebody may believe I don't enjoy / love the act of writing code.
1. I care a lot about programming, I love creating something from scratch, line by line. But: at this point, I want to do programming in a way that makes me special, compared to machines. When the LLM hits a limit, and I write a function in a way it can't compete, that is good.
2. If I write a very small program that is like a small piece of poetry, this is good human expression. I'll keep doing this as well.
3. But, if I need to develop a feature, and I have a clear design idea, and I can do it in 2 hours instead of 2 weeks, how to justify to myself that, just for what I love, I will use a lot more time? That would be too much of ego-centric POV, I believe.
4. For me too this is painful, as a transition, but I need to adapt. Fortunately I also enjoyed a lot the design / ideas process, so I can focus on that. And write code myself when needed.
5. The reason why I wrote this piece is because I believe there are still a lot of people that are unprepared for the fact we are going to be kinda of obsolete in what defined us, as a profession: the ability to write code. A complicated ability requiring a number of skills at the same time, language skills, algorithms, problem decomposition. Since this is painful, and I believe we are headed in a certain direction, I want to tell the other folks in programming to accept reality. It will be easier, this way.
I think there is a danger in the enthusiasm for AI inside of these excellent points, namely that the skills that make a good programmer are not inherent, they are learned.
The comparison would be a guy who is an excellent journeyman electrician. This guy has visual-spatial skills that makes bending and installing conduit a kind of art. He has a deep and intuitive understanding of how circuits are balanced in a panel, so he does not overload a phase. But he was not born with them. These are acquired over many years of labor and tutelage.
If AI removes these barriers--and I think it will, as AI-enhanced programmers will out-perform and out-compete those who are not in today's employment market--then the programmer will learn different skills that may or may not be in keeping with language skills, algorithms, problem decomposition, etc. They may in fact be orthogonal to these skills.
The effect of this may be an improvement, of course. It's hard to say for sure as I left my crystal ball in my other jacket. But it will certainly be different. And those who are predisposed for programming in the old-school way may not find the field as attractive because it is no longer the same sort of engineering, something like the difference between the person that designs a Lego set and the person that assembles a Lego set. It could, in fact, mean that the very best programmers become a kind of elite, able to solve most problems with just a handful of those elite programmers. I'm sure that's the dream of Google and Microsoft. However this will centralize the industry in a way not seen since perhaps IBM, only with a much smaller chance of outside disruption.
Maybe solving more trivial problems with AI will left novice programmer to do more depth problems and will make them better faster, because they will spend time solving problems that matter.
That is possible, for sure. But think of it like a person learning the piano. You could practice your arpeggios on a Steinway, or you can buy a Casio with an arpeggiator button.
At a certain point, the professional piano player can make much better use of the arpeggiator button. But the novice piano player benefits greatly from all the slogging arpeggio practice. It's certainly possible that skipping all that grunt work will improve and/or advance music, but it's hardly a sure thing. That's the experiment we're running right now with AI programming. I suppose we'll see soon enough, and I hope I'm utterly wrong about the concerns I have.
That's really interesting, but i'm wondering if this is as rational as it looks.
> we are going to be kinda of obsolete in what defined us, as a profession: the ability to write code
Is it a fact, really? I don't think "writing code" is a defining factor, maybe it's a prerequisite, as being able to write words hardly defines "a novelist".
Anyway, prompt writing skills might become obsolete quite soon. So the main question might be to know which trend of technological evolution to pick and when, in order not to be considered obsolete. A crystal ball might still be more relevant than LLMs for that.
I call it "the ability to communicate intent [using a programming language]" and suddenly building with AI looks at lot more like the natural extension of what we used to do writing code by ourselves.
I don't think our profession was writing code to begin with (and this may be a bit uuhh. rewriting history?); what we do is take an idea, requirements, an end goal and make it reality. Often times that involves writing code, but that's only one aspect of the software developer's job.
Analogy time because comment sections love analogies. A carpenter can hammer nails, screw screws, make holes, saw wood to size. If they then use machines to make that work easier, do they stop being carpenters?
It's good if not essential to be able to write code. It's more important to know what to write and when. Best thing to do at this point is to stop attaching one's self-worth with the ability to write code. That's like a novelist (more analogies) who praises their ability to type at 100wpm. The 50 shades books proved you don't need to either touch type (the first book was mostly written on a blackberry apparently) or be good at writing to be successful, lol.
Agreed - as I see it, it's akin to the transitions from machine code -> assembly language -> C -> Javascript. As time went by, knowing the deep internals of the machine became less and less necessary, even though having that knowledge still gives an engineer a useful insight into their work and often makes them better at their job. The goal remains the same - make the computer do the thing; only the mechanism changes as the tools evolve.
"-> AI" is just the next step along that journey. Maybe it will end at "-> AGI" and then humans will engage in programming mostly for the craft and the pleasure of it, like other crafts that were automated away over the ages.
As a specific example of this, the U.S. 18F team had helped the Forest Service a decade ago with implementing a requirement to help people get a permit to cut down a Christmas tree.
Although there was a software component for the backend, the thing that the actual user ended up with was a printed-out form rather than a mobile app or QR code. This was a deliberate design decision (https://greacen.com/media/guides/2019/02/12/open-forest-laun...), not due to a limitation of software.
I still really, really, really struggle to see how humans are going to maintain and validate the programs written by LLMs if we no longer know (intimately) how to program. Any thoughts?
Very few people have the expertise to write efficient assembly code, yet everyone relies on compilers and assemblers to translate high-level code to byte-level machine code. I think same concept is true here.
Once coding agents become trivial, few people will know the detail of the programming language and make sure intent is correctly transformed to code, and the majority will focus on different objectives and take LLM programming for granted.
No, that's a completely different concept, because we have faultless machines which perfectly and deterministically translate high-level code into byte-level machine code. This is another case of (nearly) perfect abstraction.
On the other hand, the whole deal of the LLM is that it does so stochastically and unpredictably.
The unpredictable part isn't new - from a project manager's point of view, what's the difference between an LLM and a team of software engineers? Both, from that POV, are a black box. The "how" is not important to them, the details aren't important. What's important is that what they want is made a reality, and that customers can press on a button to add a product to their shopping cart (for example).
LLMs mean software developers let go of some control of how something is built, which makes one feel uneasy because a lot of the appeal of software development is control and predictability. But this is the same process that people go through as they go from coder to lead developer or architect or project manager - letting go of control. Some thrive in their new position, having a higher overview of the job, while some really can't handle it.
"But this is the same process that people go through as they go from coder to lead developer or architect or project manager - letting go of control."
In those circumstances, it's delegating control. And it's difficult to judge whether the authority you delegated is being misused if you lose touch with how to do the work itself. This comparison shouldn't be pushed too far, but it's not entirely unlike a compiler developer needing to retain the ability to understand machine code instructions.
As someone that started off with assembly issues for a large corporation - assembly code may sometimes contain very similiar issues that mroe high-level code those, the perfection of the abstraction is not guaranteed.
But yeah, there's currently a wide gap between that and a stochastic LLM.
We also have machines that can perfectly and deterministically check written code for correctness.
And the stohastic LLM can use those tools to check whether its work was sufficient, if not, it will try again - without human intervention. It will repeat this loop until the deterministic checks pass.
You can make analysers that check for deeply nested code, people calling methods in the wrong order and whatever you want to check. At work we've added multiple Roslyn analysers to our build pipeline to check for invalid/inefficient code, no human will be pinged by a PR until the tests pass. And an LLM can't claim "Job's Done" before the analysers say the code is OK.
And you don't need to make one yourself, there are tons you can just pick from:
> It's not like testing code is a new thing. Junit is almost 30 years old today.
Unit tests check whether code behaves in specific ways. They certainly are useful to weed out bugs and to ensure that changes don't have unintended side effects.
> And code correctness:
These are tools to check for syntactic correctness. That is, of course, not what I meant.
Algorithmic correctness? Unit tests are great for quickly poking holes in obviously algorithmically incorrect code, but far from good enough to ensure correctness. Passing unit tests is necessary, not sufficient.
Syntactic correctness is more or less a solved problem, as you say. Doesn't matter if the author is a human or an LLM.
It depends on the algorithm of course. If your code is trying to prove P=NP, of course you can't test for it.
But it's disingenuous to claim that even the majority of code written in the world is so difficult algorithmically that it can't be unit-tested to a sufficient degree.
Suppose you're right and the "majority of code" is fully specified by unit testing (I doubt it). The remaining body of code is vast, and the comments in this thread seem to overlook that.
> Very few people have the expertise to write efficient assembly code, yet everyone relies on compilers and assemblers to translate high-level code to byte-level machine code. I think same concept is true here.
That's a poor analogy which gets repeated in every discussion: compilers are deterministic, LLMs are not.
> That's a poor analogy which gets repeated in every discussion: compilers are deterministic, LLMs are not.
Compilers are not used directly, they are used by human software developers who are also not deterministic.
From the perspective of an organization with a business or service-based mission, they already know how to supervise non-deterministic LLMs because they already know how to supervise non-deterministic human developers.
Why does it matter if LLMs are not deterministic? Who cares?
There should be tests covering meaningful functionality, as long as the code passes the tests, ie. the externally observable behaviour is the same, I don't care. (Especially, if many tests can also be autogenerated with the LLM.)
>>> Very few people have the expertise to write efficient assembly code, yet everyone relies on compilers and assemblers to translate high-level code to byte-level machine code. I think same concept is true her
>> That's a poor analogy which gets repeated in every discussion: compilers are deterministic, LLMs are not.
> Why does it matter if LLMs are not deterministic? Who cares?
In the context of this analogy, it matters. If you're not using this analogy, then sure, only the result matters. But when the analogy being used is deterministic, then, yes, it matters.
You can't very well claim "We'll compare this non-deterministic process to this other deterministic process that we know works."
The difference is that if you write in C you can debug in C. You don't have to debug the assembly. You can write an english wish list for an LLM but you will still have to debug the generated code. To debug it you will need to understand it.
> how humans are going to maintain and validate the programs written by LLMs if we no longer know (intimately) how to program
Short answer: we wouldn’t be able to. Slightly-less short answer: unlikely to happen.
Most programmers today can’t explain the physics of computation. That’s fine. Someone else can. And if nobody can, someone else can work backwards to it.
> > how humans are going to maintain and validate the programs written by LLMs if we no longer know (intimately) how to program
> Short answer: we wouldn’t be able to.
That's a huge problem! A showstopper for many kinds of programs!
> Slightly-less short answer: unlikely to happen.
Could you elaborate?
> Most programmers today can’t explain the physics of computation. That’s fine. Someone else can. And if nobody can, someone else can work backwards to it.
That's not the same at all. We have properly abstracted away the physics of computation. A modern computer operates in a way where, if you use it the way you've been instructed to, the physics underlying the computations cannot affect the computation in any undocumented way. Only a very few (and crucically, known and understood!!) physical circumstances can make the physics influence the computations. A layperson does not need to know how those circumstances work, only roughly what their boundaries are.
This is wildly different from the "abstraction" to programming that LLMs provide.
> That's a huge problem! A showstopper for many kinds of programs!
We have automated validation and automated proofs.
Proof is necessary. Do you validate the theorem prover, or trust that it works? Do you prove the compiler is correctly compiling the program (when it matters, you should, given they do sometimes re-write things incorrectly) or trust the compiler?
> We have properly abstracted away the physics of computation. A modern computer operates in a way where, if you use it the way you've been instructed to, the physics underlying the computations cannot affect the computation in any undocumented way.
You trust the hardware the code is running on? You shouldn't.
Rowhammer comes to mind, but it's hardly the only case. US banned some Chinese chips for unspecified potential that this was going on.
For some people it's OK to run a few simple tests on the chip's output to make sure it doesn't have something like the Pentium FDIV bug, for others they remove the silicon wafer from the packaging and scan it with an electron microscope, verify not just each transistor is in the right place but also that the wires aren't close enough to have currents quantum tunnelling or act as an antenna that leaks out some part of a private key.
Some people will go all the way down to the quantum mechanics. Exploits are possible at any level, domains where the potential losses exceed the cost of investigation do exist, e.g. big countries and national security.
Proof is necessary. The abstraction of hardware is good enough for most of us, and given the excessive trust already given to NPM and other package management tools, LLM output that passes automated tests is already sufficient for most.
People like me who don't trust package management tools, or who filed bugs with Ubuntu for not using https enough and think that Ubuntu's responses and keeping the bug open for years smelled like "we have a court order requiring this but can't admit it" (https://bugs.launchpad.net/ubuntu-website-content/+bug/15349...)… well, I can't speak for the paranoid, but I'm also the curious type who learned how to program just because the book was there next to the C64 game tapes.
> We have automated validation and automated proofs.
Example?
> Proof is necessary. Do you validate the theorem prover, or trust that it works? Do you prove the compiler is correctly compiling the program (when it matters, you should, given they do sometimes re-write things incorrectly) or trust the compiler?
I trust that the people who wrote the compiler and use it will fix mistakes. I trust the same people to discover compiler backdoors.
As for the rest of what you wrote: you're missing the point entirely. Rowhammer, the fdiv bug, they're all mistakes. And sure, malevolence also exists. But when mistakes or malevolence are found, they're fixed, or worked around, or at least documented as mistakes. With an LLM you don't even know how it's supposed to behave.
> you're missing the point entirely. Rowhammer, the fdiv bug, they're all mistakes. And sure, malevolence also exists.
Rowhammer was a thing because the physics was ignored. Calling it a mistake is missing the point, it demonstrates the falseness of the previous claim:
We have properly abstracted away the physics of computation. A modern computer operates in a way where, if you use it the way you've been instructed to, the physics underlying the computations cannot affect the computation in any undocumented way.
Rowhammer *is* the physics underlying the computations affecting the computation in a way that was undocumented prior to it getting discovered and, well, documented. Issues like this exist before they're documented, and by definition nobody knows how many unknown things like this have yet to be found.
> But when mistakes or malevolence are found, they're fixed, or worked around, or at least documented as mistakes.
If you vibe code (as in: never look at the code), then find an error with the resulting product, you can still just ask the LLM to fix that error.
I only had a limited time to experiment with this before Christmas (last few days of a free trial, thought I'd give it a go to see what the limits were), and what I found it doing wrong was piling up technical debt, not that it was a mysterious ball of mud beyond its own ability to rectify.
> With an LLM you don't even know how it's supposed to behave.
LLM generated source code: if you've forgotten how to read the source code it made for you to solve your problem and can't learn how to read that source code and can't run the tests of that source code, at which point it's as interpretable as psychology.
The LLMs themselves: yes, this is the "interpretability" problem, people are working on that.
I am of course aware. Any malevolent backdoor in your compiler could also exist in your LLM. Or the compiler that compiled the LLM. So you can never do better.
> Rowhammer is the physics underlying the computations affecting the computation in a way that was undocumented prior to it getting discovered and, well, documented. Issues like this exist before they're documented, and by definition nobody knows how many unknown things like this have yet to be found.
Yep. But it's a bug. It's a mistake. The unreliability of LLMs is not.
> If you vibe code (as in: never look at the code), then find an error with the resulting product, you can still just ask the LLM to fix that error.
Of course. But you need skills to verify that it did.
> LLM generated source code: if you've forgotten how to read the source code it made for you to solve your problem and can't learn how to read that source code and can't run the tests of that source code, at which point it's as interpretable as psychology.
Reading source code is such a minute piece of the task of understanding code that I can barely understand what you mean.
At the current time this is essentially science fiction though. This something that the best funded companies on the planet (as well as many many others) work on and seem to be completely unable to achieve despite trying their best for years now, despite an incredible hype.
It feels like if those resources were poured in nuclear fusion for example we'd have it production ready by now.
The field is also not a couple of years old, this has been tried for decades. Sure only now companies decided to put essentially "unlimited" resources into it, but while it showed that certain things are possible and work extremely well, it also strongly hinted that at least the current approach will not get us there, especially not without significant trade-off (that whole over training vs "creativity" and hallucination topic).
Doesn't mean it won't come, but that it doesn't appear a "we just need a bit more development" topic. The state hasn't changed much. Models became bigger and bigger and people added that "thinking" hack and agents and agents for agents, but it also didn't change much about the initial approach and its limitations, given that they haven't cracked these problems after years of hyped funding.
Would be amazing if we would have AIs that automate research and maybe help us fix all the huge problems the world is facing. I'd absolutely love that. I'd also love it if people could easily create tools, games, art. However that's not the reality we live in. Sadly.
Tutoring – whether AI or human – does not provide the in-depth understanding necessary for validation and long-term maintenance. It can be a very useful step on the way there, but only a step.
Same how we do it now - look at the end result, test it. Testers never went away.
Besides, your comment goes by the assumption that we no longer know (intimately) how to program - is that true? I don't know C or assembly or whatever very well, but I'm still a valuable worker because I know other things.
I mean it could be partially true - but it's like having years of access to Google to quickly find just what I need, meaning I never learned how to read e.g. books on software development or scientific paper end to end. Never felt like I needed to have that skill, but it's a skill that a preceding generation did have.
> Besides, your comment goes by the assumption that we no longer know (intimately) how to program - is that true? I don't know C or assembly or whatever very well, but I'm still a valuable worker because I know other things.
The proposal seems to be for LLMs to take over the task of coding. I posit that if you do not code, you will not gain the skills to do so well.
> I mean it could be partially true - but it's like having years of access to Google to quickly find just what I need, meaning I never learned how to read e.g. books on software development or scientific paper end to end.
I think you've misunderstood what papers are for or what "the previous generation" used them for. It is certainly possible to extract something useful from a paper without understanding what's going on. Googling can certainly help you. That's good. And useful. But not the main point of the paper.
Fair question but haven't we been doing this for decades? Very few people know how to write assembly and yet software has proliferated. This is just another abstraction.
> Fair question but haven't we been doing this for decades? Very few people know how to write assembly and yet software has proliferated. This is just another abstraction.
Not at all. Given any "layperson input", the expert who wrote the compiler that is supposed to turn it into assembly can describe in excruciating detail what the compiler will do and why. Not so with LLMs.
Said differently: If I perturb a source code file with a few bytes here and there, anyone with a modicum of understanding of the compiler used can understand why the assembly changed the way it did as a result. Not so with LLMs.
But there's a limit to that. There's (relatively) very few people that can explain the details of e.g. a compiler, compared to for example React front-end developers that build B2C software (...like me). And these software projects grow, ultimately to the limit of what one person can fit in their head.
Which is why we have lots of "rules" and standards on communication, code style, commenting, keeping history, tooling, regression testing, etc. And I'm afraid those will be the first to suffer when code projects are primarily written by LLMs - do they even write unit tests if you don't tell them to?
Really enjoyed your article, and it reflects a lot of the pain-points I experience with models. I tend to still write and review LLM code after creation, but there is definitely a shift within how much code I create "artisinally" and how much is reviewd in terms of scale.
If I need to implement a brand new feature for the project, I will find myself needing to force a view into a LLM because it will help me achieve 80% of the feature in 1% of the time, even if the end result requires a scale of refactoring, it's rarely the time that the original feature would've taken me.
But, I think that's also because I have solid refactoring foundations, I know what makes good code, and I think if I had access to these tools 5 years ago, I would not be able to achieve that same result, as LLMs typically steer towards junior level coding as a consequence of their architecture.
This is by far the best summary of the state of affairs, or rather, the most sensible perspective that one should have on the state of affairs, that I've read so far.
Of the four coding examples you describe, I find none of them compelling either in their utility or as a case for firing a dev (with one important caveat [0]).
In each example, you were already very familiar with the problem at hand, and that probably took far longer than any additional time savings AI could offer.
0. Perhaps I consider your examples as worthless simply because you gloss over them so quickly, in which case that greatly increases the odds in most companies that you would be fired.
All of that makes a lot of sense. And unlike a lot of both pro-AI and anti-AI people I would find it great if it was the case. Unlike maybe a lot of people here I am less attached to this profession as a profession. I'd also love it if I could have some LLM do the projects I always wanted to finish. It would be essentially Christmas.
However your experiences really clash with mine and I am trying to work out why, because so far I haven't been able to copy your workflow with success. It would be great if I could write a proper spec and the output of the LLM would be good (not excellent, not poetry, but just good). However the output for anything that isn't "stack overflow autocomplete" style it is abysmal. Honestly I'd be happy if good output is even on the horizon.
And given that "new code" is a lot better than working on an existing project and an existing LLM generated project being better than a human made project and it still being largely bad, often with subtle "insanity" I have a hard time to apply what you say to reality.
I do not understand the disconnect. I am used to writing specs. I tried a lot of prompting changes, to a degree where it almost feels like a new programming language. Sure there are things that help, but the sad reality is that I usually spend more time dealing with the LLM than I'd need to write that code myself. And worse still, I will have to fix it and understand it, etc. to be able to keep on working on it and "refining" it, something that simply isn't needed at least to that extent if I wrote that code myself.
I really wished LLMs would provide that. And don't get me wrong, I do think there are really good applications for LLMs. Eg anything that needs a transform where even a complex regex won't do. Doing very very basic stuff where one uses LLMs essentially as an IDE-integrated search engine, etc.
However the idea that it's enough to write a spec for something even semi-novel currently appears to be out of reach. For trivial generic code it essentially saves you from either writing it yourself copy pasting it off some open source projects.
Much context, for the question that hopefully explains a lot of stuff. Those 2 hours that you use instead of two weeks. How do you spend them? Is that refining prompts, is that fixing the LLM output, is that writing/adapting specs, is it something else?
Also could it be that there is a bias on "time spent" because of it being different work or even just a general focus on productivity, more experience, etc.?
I am trying to understand where that huge gap in experience that people have really stems from. I read your posts, I watch video on YouTube, etc. I just haven't seen "I write a spec [that is is shorter/less effort than the actual code] and get good output". Every time I read claims about it in blog posts and so on there appear to be parts missing to reproduce the experience.
I know that there are a lot of "ego-centric POV" style AI "fear". People of course have worries about their jobs, and I understand. However, personally I really don't and as mentioned I'd absolutely love to use it like that on some projects, but whenever I try to replicate experiences that aren't just "toying" in the sense of anything that even has basic reliability requirements and is a bit more complex I fail to do so and it's probably me, but I tried for at least a year to replicate such things and it's failure after failure even for more simple things.
That said there are productivity gains with autocomplete, transforming stuff and what people largely call "boilerplate" as well as more quickly writing small helpers that I'd otherwise have copied off some older project. Those things work good enough, just like how autocomplete is good enough. For bigger and more novel things where a search engine is also not the right approach it fails, but this is where the interesting bits are. Having topics that haven't been solved a hundred times over.
Thanks for the post. I found it very interesting and I agree with most of what you said. Things are changing, regardless of our feelings on the matter.
While I agree that there is something tragic about watching what we know (and have dedicated significant time and energy in learning) devalued. I'm still exited for the future, and for the potential this has. I'm sure that given enough time this will result in amazing things that we cannot even imagine today. The fact that the open models and research is keeping up is incredibly important, and probably the main things that keeps me optimistic for the future.
I wanted to provide some more context that is not part of the blog post. Since somebody may believe I don't enjoy / love the act of writing code.
1. I care a lot about programming, I love creating something from scratch, line by line. But: at this point, I want to do programming in a way that makes me special, compared to machines. When the LLM hits a limit, and I write a function in a way it can't compete, that is good.
2. If I write a very small program that is like a small piece of poetry, this is good human expression. I'll keep doing this as well.
3. But, if I need to develop a feature, and I have a clear design idea, and I can do it in 2 hours instead of 2 weeks, how to justify to myself that, just for what I love, I will use a lot more time? That would be too much of ego-centric POV, I believe.
4. For me too this is painful, as a transition, but I need to adapt. Fortunately I also enjoyed a lot the design / ideas process, so I can focus on that. And write code myself when needed.
5. The reason why I wrote this piece is because I believe there are still a lot of people that are unprepared for the fact we are going to be kinda of obsolete in what defined us, as a profession: the ability to write code. A complicated ability requiring a number of skills at the same time, language skills, algorithms, problem decomposition. Since this is painful, and I believe we are headed in a certain direction, I want to tell the other folks in programming to accept reality. It will be easier, this way.