AI Is Prompting an Evolution, Not Extinction, for Coders (thestar.com.my) 71
AI coding assistants are reshaping software development, but they're unlikely to replace human programmers entirely, according to industry experts and developers. GitHub CEO Thomas Dohmke projects AI could soon generate 80-90% of corporate code, transforming developers into "conductors of an AI-empowered orchestra" who guide and direct these systems.
Current AI coding tools, including Microsoft's GitHub Copilot, are delivering 10-30% productivity gains in business environments. At KPMG, developers report saving 4.5 hours weekly using Copilot, while venture investment in AI coding assistants tripled to $1.6 billion in 2024. The tools are particularly effective at automating routine tasks like documentation generation and legacy code translation, according to KPMG AI expert Swami Chandrasekaran.
They're also accelerating onboarding for new team members. Demand for junior developers remains soft, however, though analysts say it's premature to attribute this directly to AI adoption. Training programs like Per Scholas are already adapting, incorporating AI fundamentals alongside traditional programming basics to prepare developers for an increasingly AI-augmented workplace.
Current AI coding tools, including Microsoft's GitHub Copilot, are delivering 10-30% productivity gains in business environments. At KPMG, developers report saving 4.5 hours weekly using Copilot, while venture investment in AI coding assistants tripled to $1.6 billion in 2024. The tools are particularly effective at automating routine tasks like documentation generation and legacy code translation, according to KPMG AI expert Swami Chandrasekaran.
They're also accelerating onboarding for new team members. Demand for junior developers remains soft, however, though analysts say it's premature to attribute this directly to AI adoption. Training programs like Per Scholas are already adapting, incorporating AI fundamentals alongside traditional programming basics to prepare developers for an increasingly AI-augmented workplace.
Let's talk about unbiased sources (Score:5, Informative)
GitHub CEO Thomas Dohmke answers to the Microsoft Board, who owns GitHub. That same board also has one of the most expensive investments in a tech company, ever, through its demi-acquisition of OpenAI.
Does this strike the Author of The Star's article, Steve Lohr, as a conflict of interest worth noting with as much as a small half sentence at the end of the article? No. Why would it? Bloviating about the future of technology is what tech CEOs do, and questioning them might imperial your access.
Re:Let's talk about unbiased sources (Score:4, Funny)
Re: (Score:3)
Help, help! I'm being imperialed!
Re: (Score:3)
By whom? Vlad the Imperialer?
Re: (Score:2)
GitHub CEO Thomas Dohmke projects AI could soon generate 80-90% of corporate code,
Any AI prediction that talks about the future is nonsense. In this case, marketing nonsense. The only things that matter are current capabilities.
That is how to hype/noise from reality.
Evolution might be optimistic (Score:5, Insightful)
If anything, it will probably keep us frozen in time. I highly doubt that it will spur any meaningful evolution, and innovation is probably likely to suffer as well from the added maintenance workload that more code generates.
Re: (Score:2)
If anything, it will probably keep us frozen in time. I highly doubt that it will spur any meaningful evolution, and innovation is probably likely to suffer as well from the added maintenance workload that more code generates.
Lets hope at the very least it can read already-written code and make it more calculation and memory efficient. It'd be great if the next was 10% more efficient.
Re: (Score:2)
If anything, it will probably keep us frozen in time. I highly doubt that it will spur any meaningful evolution, and innovation is probably likely to suffer as well from the added maintenance workload that more code generates.
Hmmm. You are trying to hide your denialism behind a thin veneer of skepticism. The idea that AI will “freeze” development instead of accelerating it ignores every historical precedent we have in software and technology. From compilers to IDEs to version control, every advance that has streamlined coding has led to more innovation, not less.
You say innovation will suffer due to “the added maintenance workload that more code generates.” That is a bizarre inversion of reality. AI-ass
Re: (Score:2)
I haven't seen big advances in software in over a decade. Honestly, the low-code no-code mentality, along with taboos to write original code when the internet should be scoured instead has really made a lot of software utterly stupid. Bloated, slow, web based nonsense by people who have minimum training for minimum pay.
If a programmer can make use of AI for coding (*) they likely are already just scouring the web for code snippets, or else they do duct-tape programming of tying together other peoples' mod
Re: Evolution might be optimistic (Score:2)
> You say innovation will suffer due to âoethe added maintenance workload that more code generates.â That is a bizarre inversion of reality. AI-assisted development reduces boilerplate, flags issues earlier, and improves documentationâ"precisely the things that reduce maintenance burdens.
You are assuming AI does everything correctly all the time. It simply does not. When it screws up - which is often, and sometimes deceptively so - it costs time and resources. Even when it does help, you
Only a threat to reference books ... (Score:2)
Rather than stand up, walk to my bookshelf, and grab my data structures textbook from college. I ask the AI what sorting algorithm works best with data of some particular nature and it gives me a brief text paragraph describing the relevant algorithm(s) and some sample code that can be used as a template. Maybe I can then go wild and say given a class or structure with member x, show that algorithm sorting by x.
That seems to be where AI
we gonna give-a you cacciatore (Score:3, Insightful)
Companies don't need programmers to create code. They need programmers to maintain code. From the anecdotes I've read about LLM code generation, it's all either mindless boilerplate, the kind of thing that anybody could type up in half an hour that's too simple to even have errors; or, for more complicated things, bad, unmaintainable code. The LLM won't really buy you any time writing boilerplate, because it all has to be well-specified beforehand and checked afterwards anyway. And if you use it to write things more complicated than boilerplate, it will make spaghetti out of it. And it can't then fix the spaghetti, because every time you turn it loose on an existing codebase, it just makes more spaghetti. Sooner or later, the spaghetti assumes critical mass, and you can't fix one thing without breaking three others.
It's basically just a machine for automating the taking on of technical debt.
Re:we gonna give-a you cacciatore (Score:5, Insightful)
so you haven't used LLM code generation, that's ok
wait, how can you opine on something you have never used or seen, and only "read about"?
I HAVE LOUD OPINIONS ABOUT THINGS I DON'T UNDERSTAND.
Re: (Score:1)
So in your view, atheists aren't allowed to have views on god?
As an atheist myself, I don't think I'm qualified to share my views on god or gods.
It's like me trying to have an opinion on A Game of Thrones. I never watches the show or read the book (past the first chapter). My personal opinion is I'm not interested in it. But sharing my views on various aspects of whatever religion we talk about is kind of pointless, since I don't believe it, I don't spend time obsessing about it, and the fanbase doesn't wasn't to hear it from me.
Re: (Score:2, Insightful)
By that logic, religious people aren't allowed to have views on god either, since they have never met such a being but have only read about him.
Re: (Score:1)
Re: (Score:1)
"so you haven't used LLM code generation, that's ok"
5 mins with ChatGPT will demonstrate that it can't even get the boilerplate basics correct a lot of the time, never mind generating application specific code.
Re: (Score:2)
It's ok if you want to believe that. You can lie to yourself all you want but that doesn't really change facts.
The reality is that it generates good boilerplate code.
Re: we gonna give-a you cacciatore (Score:2)
Not for C++ it doesnt. Its not even close. The code ie actually laughable sometimes.
Re: (Score:1)
Read about foxes and hedgehogs. Foxes usually win.
The most famous experiment was about predicting what would happen to Soviet Union over the next 10 years (this was 1980s). Two separate teams were created to make predictions. Hedgehogs, who were experts on the Soviet Union (military, economy, social, etc). IOW domain specialists. And foxes, who were general
Re: (Score:1)
I HAVE LOUD OPINIONS ABOUT THINGS I DON'T UNDERSTAND.
Yeah, it's crazy. All these people confidently telling me that tools I use don't work. When I know that they do, because I actually use them.
Re: (Score:2)
Couldn't have said it better myself. Productivity measured how? None of these statistics and 10-30% is pretty bad if it's just productivity in "generating code" - slamming out code is usually the wrong choice. A good senior developer makes strategic decisions that improve the maintainability of code and fix problems that nobody's even reported yet, AI won't do that.
Re: (Score:2)
Consider that the entire post should be met with skepticism, as its premise is fundamentally co-opted by its source.
Microsoft first poised a developer program, and in doing so, made Windows a platform for making money. Microsoft's Developer Network rivaled most at the time, and grew to out-shadow any of them. From Oracle to Apple, their money-making, Windows-advancing network made money for Microsoft and developers. Developers sold and integrated their wares to the organized world.
Now the Microsoft Buzz is
Re: (Score:1)
Re: (Score:2)
The LLM won't really buy you any time writing boilerplate, because it all has to be well-specified beforehand and checked afterwards anyway. And if you use it to write things more complicated than boilerplate, it will make spaghetti out of it. And it can't then fix the spaghetti, because every time you turn it loose on an existing codebase, it just makes more spaghetti. Sooner or later, the spaghetti assumes critical mass, and you can't fix one thing without breaking three others.
They only need to be good enough to sell into businesses as middleman-ware. Any damage they do will be solved (at a price) by the next generation of LLMs after all of the hapless customers train them.
So Ars technica has a pretty good article (Score:2, Interesting)
In the old days when we still had competition you could get a bunch of new tech from new competitors. That doesn't happen anymore. Market consolidation and a complete lack of antitrust law enforcement means that if you try to go fast and break things you will either get run out of business or if you're really
Re: (Score:1)
Stop spamming this site! You keep on posting the same thing in a loop which makes it spam. You will be banned for spamming. Also, you are using alt accounts to manipulate the moderation system. You are defacing this site with your spam and other shenanigans:
You posted the same not even 24 hours ago, see first item in references below!
Market consolidation and a complete lack of antitrust law enforcement means that if you try to go fast and break things you will either get run out of business or if you're really really lucky a few million bucks tossed your way to go away.
references:
https://slashdot.org/comments.... [slashdot.org]
I suspect if we had proper antitrust law enforcement HP would not be allowed to do these kind of sweetheart up front deals that they
Re:we gonna give-a you cacciatore (Score:4, Interesting)
Hmmm. All of the points you are asserting have been made by others -- mostly by software engineers who are in a state of denial about where AI is taking the industry. The actual industry results show that AI coding tools improve productivity, reduce maintenance overhead, and are rapidly evolving beyond simple boilerplate generation. Dismissing AI as a "spaghetti generator" ignores both empirical data and the direction the field is heading. Coders who embrace AI will adapt and thrive—those who refuse to evolve will be left behind.
Companies don't need programmers to create code. They need programmers to maintain code.
False dichotomy. Companies need programmers to do both. AI-assisted coding tools are already improving the creation of new software while reducing the maintenance burden by generating better documentation, flagging errors earlier, and automating code refactoring. The article explicitly mentions that AI helps developers generate documentation and modernize legacy code, making maintenance easier, not harder.
From the anecdotes I've read about LLM code generation, it's all either mindless boilerplate, the kind of thing that anybody could type up in half an hour that's too simple to even have errors; or, for more complicated things, bad, unmaintainable code.
This is an outdated generalization. The article provides real-world examples of AI improving productivity between 10% and 30% for developers in business settings, showing that it is not limited to boilerplate. AI-assisted coding is already being used effectively at major companies like KPMG and Microsoft, and its ability to generate structured, readable code is improving rapidly.
The LLM won't really buy you any time writing boilerplate, because it all has to be well-specified beforehand and checked afterwards anyway.
This contradicts actual industry results. Developers using GitHub Copilot at KPMG save 4.5 hours per week on average, meaning that AI is indeed saving time even with verification. The reality is that AI dramatically speeds up writing and reviewing boilerplate code, freeing human developers for more complex tasks.
And if you use it to write things more complicated than boilerplate, it will make spaghetti out of it. And it can't then fix the spaghetti, because every time you turn it loose on an existing codebase, it just makes more spaghetti.
Blatantly false. AI is already being used to modernize legacy code and automate software translation into modern languages—precisely the opposite of generating "spaghetti." The article cites examples of AI improving software quality and reducing maintenance overhead, showing that with proper use, it reduces technical debt rather than increasing it.
Sooner or later, the spaghetti assumes critical mass, and you can't fix one thing without breaking three others.
This assumes developers blindly accept AI-generated code without oversight, which is neither true nor how AI is used in professional settings. Developers act as conductors of an AI-powered workflow, ensuring quality control. AI is a tool, not a replacement, and it enhances developer productivity when used correctly.
It's basically just a machine for automating the taking on of technical debt.
Incorrect. AI coding assistants help reduce technical debt by automating documentation, flagging bugs earlier, and assisting with refactoring. The article highlights how AI assists new developers in onboarding faster, which improves team efficiency.
Re: (Score:2)
10-30% matches generation of boilerplate well. And none of your examples actually support that what is generated isn't spaghetti. You're simply asserting that it's not what is happening.
Also, you're placing several limits on what you count, without being very explicit about it. "When used correctly" is something you tack on to one of your sentences, but it's central to every point you make - and you do not qualify nor quantify that.
When generators are not used correctly, which will be in the majority of cas
Re: (Score:2)
Companies don't need programmers to create code. They need programmers to maintain code. From the anecdotes I've read about LLM code generation, it's all either mindless boilerplate, the kind of thing that anybody could type up in half an hour that's too simple to even have errors; or, for more complicated things, bad, unmaintainable code. The LLM won't really buy you any time writing boilerplate, because it all has to be well-specified beforehand and checked afterwards anyway. And if you use it to write things more complicated than boilerplate, it will make spaghetti out of it. And it can't then fix the spaghetti, because every time you turn it loose on an existing codebase, it just makes more spaghetti. Sooner or later, the spaghetti assumes critical mass, and you can't fix one thing without breaking three others.
It's basically just a machine for automating the taking on of technical debt.
You are so feaking out of touch.
Doesn't sound like you're not a coder; or have coded recently with LLM assistance.
Really? (Score:2)
AI Is Prompting an Evolution, Not Extinction, for Coders
What changed? Was somebody beaten with a clue stick? I have lost count of the number of mercurial and supposedly autistic (what is it with half of these assholes claiming to be autistic anyway?) US based AI business geniuses proclaiming that software engineers and programmers will be unnecessary within two to three years and all need for human labour will end by 2030. Quite frankly the statement that "... AI could soon generate 80-90% of corporate code, transforming developers into conductors of an AI-empow
Re: (Score:2, Interesting)
what is it with half of these assholes claiming to be autistic anyway?
1 in 50-ish children in the US. Diagnoses are rising as awareness increases.
About 2% of the world's population (~10% US pop) have green eyes.
Just tell "those assholes" that autism is normal, about as common as having green eyes. Then walk away and let them implode.
Re: (Score:1)
... transforming developers into conductors of an AI-empowered orchestra
...of cats with a laser focus /s
Yeah, right. (Score:3)
Evolution toward extinction (Score:3)
According to the immediately adjacent slashdot article [slashdot.org], that "evolution" is an evolution toward fewer coders.
Re: (Score:2)
Headline must be meant for people unaware that rapid evolution is basically a lot of dead specimens and a few I've Got Mine.
All those zombie virus movies that showed the immune survivors in hovels and compounds are An Evolution Not An Extinction.
Re: (Score:2)
Best comment thus far. And that's what management certainly wants from AI.
Re: (Score:2)
Which would be a good thing. Most coders suck and probably destroy more value than they create.
Breaking News! (Score:2)
Re: (Score:2)
What makes you think this article was written by an actual reporter?
Oh, right: If it was written by AI it would probably contain a clue or two.
marching off a cliff (Score:3)
It's cannibalistic. The AI methods reduce your brain power. Sooner or later you end up with a Programming Guild that does the technology and everyone else is a drooling moron, paying thru the nose to rent the tech. The best tech will be kept proprietary. Like a bottle of 30 year old scotch, you don't bring it out at a party.
Re: (Score:3)
Yes. It is also deadly, because non-fixed tech debt will eventually kill you.
Just another tool (Score:5, Informative)
It's just another tool for developers. I originally wrote software in a text editor in one shell, compiled in another shell, and ran my program in a third. When IDEs and interactive debuggers came along it was a massive increase in efficiency of software development, but it did not lead to a sudden drop in the demand for software developers because we were more efficient.
The same can be said about every incremental step along the way (syntactic coloring, linters, automatic code deployment, on and on).
What AI is replacing for me, and speeding up my processes, is essentially in digging through documentation, API reference, and code examples, to more succinctly answer questions I may have.
A perfect example - I recently decided to re-implement some custom software of mine (a pretty decent sized Android application) to Qt so I could run it on many more devices (specifically on Raspberry Pis). ChatGPT and Gemini proved extremely valuable getting me off the ground and answering questions, vastly increasing my performance compared to dredging through documentation. I could ask questions about the pros and cons of using various APIs / paradigms within Qt and make much faster decisions with that distilled information.
Instead of looking at hello world examples, or trying to find the example projects that most closely exhibited the functionality I needed, I could ask for specific bits of code to be generated.
Still, there were a lot of missteps and incorrect answers provided by AI (it trying to use APIs, etc, that were deprecated or flat out mismatching versions of Qt was a big one I constantly battled, even when I explicitly told it to use version 6.8), but all in all I was able to produce that software in probably 1/2 to 1/3 the time prior to AI.
So it was just another tool that made my work more efficient, and in a very specific way it reduced the learning curve and got me up to speed in a large library / app framework I was not yet familiar with. In other words AI made me learn much faster than I would have the "old fashioned" way of pouring through documentation and code examples unrelated to the specific tasks I was trying to solve.
Re:Just another tool (Score:4)
"I originally wrote software in a text editor in one shell, compiled in another shell, and ran my program in a third"
Thats now I still develop. I just find IDEs get in my way and their text editors don't hold a candle to vim for fast editing large chunks of code.
Re: (Score:2)
Me too, the biggest change for me is going from having to do Ctrl-Alt-Fn to switch windows to moving the mouse between xterms.
But an LSP integrated with vim is pretty much essential IMO. While I still use grep a lot, the LSP makes many things much more efficient.
Re: (Score:2)
Fine if you have access to an LSP. Plus the compiler works for me for syntax checking. Fixing syntatic bugs is not where my time is spent.
Re: (Score:1)
So it was just another tool that made my work more efficient, and in a very specific way it reduced the learning curve and got me up to speed in a large library / app framework I was not yet familiar with. In other words AI made me learn much faster than I would have the "old fashioned" way of pouring through documentation and code examples unrelated to the specific tasks I was trying to solve.
Exactly. You get it.
Re: Just another tool (Score:2)
This might one of the most useful comments about AI in coding I've ever read.
My programming is mostly in R, so it's much less common for me to need master an extensive framework or API.
That's why I've yet to find a case where I can even formulate a prompt for LLM code generation.
One thing I sure wish someone would build an AI-ish thing to do is extract the cardinality of a dataset I find, plus report on the nature of missingness. That's a time-consuming pain that I'd be happy to have automated.
It Did What? (Score:3)
"Wait, our software is generating false reports? Mismanaging payroll? Said our firm worships Hitler? Fix it! Find the bug in the code! Wait, that's not how it works? No one knows how it works? Um, ...."
Re: (Score:2)
Oooops....
Although, Hitler is so yesterday. Now, if you worship Putin and Xi, then we are getting somewhere.
Re: (Score:1)
Domain-clueless DOGE at work.
The problem is if you're on slashdot (Score:2)
And there's nothing wrong with that. Just because you're not one of the top
One of the things we never really think about because we are born into it is that the way things ar
Re: (Score:2)
Then you are probably not the kind of cutting-edge genius that's going to survive all the automation that's coming.
Speak for yourself ;-)
That said, I would estimate that 80-90% of all coders are bad ones and need to fear AI. The rest, not so much. Or at all.
Re: (Score:2)
Wild Guessing (Score:2)
GitHub CEO Thomas Dohmke projects AI could soon generate 80-90% of corporate code...
Hype alert! Github CEO says the Github AI is great! I guess he has to sell the dream to make his bonuses, but 80-90%, really? Has he realized he's not living in the 24th century Star Trek universe?
Current AI coding tools, including Microsoft's GitHub Copilot, are delivering 10-30% productivity gains in business environments.
I'd really love to know how they measure that. My guess is that it's a "wild guess" [or a WAG], unless they really had 2 people of comparable skill do the same task 1 with and 1 without the AI tool, and then did that on several non-trivial tasks to truly measure.
At KPMG, developers report saving 4.5 hours weekly using Copilot...
Same question as above, although this is a little cl
AI Propganda and Lies (Score:2)
AI....Wonderfull backdoor. (Score:1)
Preceding Headline disagrees. (Score:3)
Software Engineering Job Openings Hit Five Year Low immediately followed by this story? Something seems fishy. Day after day after week we see layoffs and job cuts for coding and engineering positions, and non-stop AI hype. I believe their evolution may not be an extinction, as there will still be some coders left. But it's certainly a culling. And trying to side-step it by prettying up some marketing speak isn't going to change that.
AI is very good, but not great at programming (Score:2)
I've been programming computers in one capacity or another for decades. I've done everything from Apple Basic to Visual Basic, with C, C++, Objective C, Java, HTML/CSS/JavaScript, Python, PHP & MySQL in between, plus several others that don't merit mention here.
I have to do some programming in my job, but it is only a small part. I administer several different cloud services, so I have to deal with a lot of APIs. Sometimes the particular action I'm looking to perform is not intuitive. It is much easier
Re: (Score:2)
... and I've caught it a few times in off by 1 errors.
Yuck! These can result in exploitable vulnerabilities and reliability problems.
Only for the bad coders (Score:2)
It makes them even more of a drain on company resources. That's why we fire anyone caught using AI code generators.
Re: (Score:2)
Indeed. And good approach. It can also leak your code, put in vulnerabilities and make code exceptionally hard to maintain.
The new jobs... (Score:3)
Will be hiring coders to come in and fix the absolute mess of code that Ai is spitting out.
Re: (Score:2)
Will be hiring coders to come in and fix the absolute mess of code that Ai is spitting out.
I have found the opposite to be true.
Just today AI was giving me horrible answers. I was cursing at it. I took a short walk and then just realized that there was a horrendous bug there from an edge case and AI was trying to work around it.
I fixed that bug and the solution that LLM generated was 100% correct and even elegant.
LLMs write amazing unit tests. If you change something, you can ask it to update the unit tests as well. And, nobody's ego is hurt when you throw away unit tests that are not useful anym
Re: (Score:2)
I've been using LLMs to code for the past two years. I know full well their utility. But I also know VERY well how flawed they are and why those flaws won't ever be fixed. I recently was using Chat GPT to do a refactor of one of my personal Python tools. It worked great for modernizing my code to follow current PEP, writing docstrings, and helping make my code more readable. But you have to keep the code your working on on the smallish side because even a function that's a dozen lines long Chat GPT wil
Not that "business code" was ever difficult.... (Score:3)
The actual work there is in the interfaces, data formats, software architecture and understanding what the customer wants. The actual coding is pretty trivial.
Annoying (Score:2)
Every time some CEO pushes their shallow agenda, it's annoying.
The 10-30% productivity gain is real. Without a doubt it helps write code, especially with unfamiliar libraries.
"80-90% of all corporate code" is continuing to push a story that is highly unrealistic.
Fix bad factoring & standards instead (Score:1)
Wouldn't need much of this if frameworks were not so bloated and redundant. Factor them better and/or get a GUI-friendly web standard instead of DOMshit and maybe we wouldn't need bot grunt-help with so much busywork.
Automating stupidity just multiplies stupidity, not remove it. Smart people find a way to stop getting smacked in the face. Stupid people automate the smacking because they get used to it.
Re: (Score:2)