Is AI Impacting Which Programming Language Projects Use? (github.blog)
- Reference: 0180845232
- News link: https://developers.slashdot.org/story/26/02/23/0732245/is-ai-impacting-which-programming-language-projects-use
- Source link: https://github.blog/ai-and-ml/generative-ai/how-ai-is-reshaping-developer-choice-and-octoverse-data-proves-it/
They point to this as proof that "AI isn't just speeding up coding. It's reshaping which languages, frameworks, and tools developers choose in the first place."
> Eighty percent of new developers on GitHub use Copilot within their first week. Those early exposures reset the baseline for what "easy" means. When AI handles boilerplate and error-prone syntax, the penalty for choosing powerful but complex languages disappears. Developers stop avoiding tools with high overhead and start picking based on utility instead.
>
> The language adoption data shows this behavioral shift:
>
> — TypeScript grew 66% year-over-year
> — JavaScript grew 24%
> — Shell scripting usage in [2]AI-generated projects jumped 206%
>
> That last one matters. We didn't suddenly love Bash. AI absorbed the friction that made shell scripting painful. So now we use the right tool for the job without the usual cost.
"When a task or process goes smoothly, your brain remembers," they point out. "Convenience captures attention. Reduced friction becomes a preference — and preferences at scale can shift ecosystems."
"AI performs better with strongly typed languages. Strongly typed languages give AI much clearer constraints..."
"Standardize before you scale. Document patterns. Publish template repositories. Make your architectural decisions explicit. AI tools will mirror whatever structures they see."
"Test AI-generated code harder, not less."
[1] https://github.blog/ai-and-ml/generative-ai/how-ai-is-reshaping-developer-choice-and-octoverse-data-proves-it/
[2] https://github.blog/news-insights/octoverse/typescript-python-and-the-ai-feedback-loop-changing-software-development/
Interesting Summary (Score:3)
The style and presentation of the summary, presumably quoted from the "article" actually reads like AI-generated slop text itself. This would be very on-brand for the AI bros.
Re: (Score:2)
The claims also seem a bit sus. "Eighty percent of new developers on GitHub use Copilot within their first week." Is this the same statistic someone was debunking recently where anyone who had done something really basic (it might have been using the search facility?) was counted as "using Copilot"? A lot of organisations seem to be cautious about using code generated by AIs, or even imposing a blanket ban, so things must be very different in other parts of the industry if that 80% is also representative of
Re: (Score:2)
Microsoft is rebranding every one of their products "Copilot", so at this point it seems hard to avoid using it.
Re: Interesting Summary (Score:3)
Any organization that imposes a blanket ban on AI tools will soon be left in the dust.
To use a tautology, AI is good at what AI is good for: documentation, research, incremential coding, performance/storage tradeoff evaluation.
It is not (yet) good at architecture design or efficiency, nor even following DRY principles. It is nonetheless really, really useful for what it does well.
Re: (Score:1)
Came here to say this.. the wording is 100% AI.
my experience too (Score:4, Insightful)
"AI performs better with strongly typed languages. Strongly typed languages give AI much clearer constraints..."
As Guido van Rossum, the creator of Python once wrote,
"I've learned a painful lesson, that for small programs dynamic typing is great. For large programs, you have to have a more disciplined approach. And it helps if the language actually gives you that discipline, rather than telling you, 'Well, you can do whatever you want.'"
Re: (Score:2)
It's one of the things I like about C -- it enforces these things, but still allows you to explicitly ignore them. It would be harder to write the [1] Fast inverse square root [wikipedia.org] without it.
[1] https://en.wikipedia.org/wiki/Fast_inverse_square_root
Re: (Score:2)
Yes, C is a brilliant language.
Re: (Score:2)
I won't argue with Guido van Rossum, but in my meagre experience, a Python project is "one thing", and so yes, stronger typing would likely be useful. However, a Javascript (or TS) project, in modern frameworks is a bit like loads of small projects. That is, each component of a page, or perhaps each page is like a separate project in a way - as such, the typing problem is much smaller, and so JS or TS will be fine there (and it's possible to mix the two).
That said, AI is popularist. If TS is the predominant
Re: (Score:2)
Hi - thanks for this thoughtful response.
I do like the flexibility of Javascript-like languages - e.g. that one can just add an attribute to any object. But one pays for that:
I am working on a pure Javascript front end with a Java back end at the moment, and whenever I have to make changes to an object that needs to pass between the two, it is a real headache on the Javascript side - there is no compiler to find all the places where I used the object based on its old definition...
So pros and cons.
Needs strongly typed languages? (Score:1)
So like low-skill coders then? Figures.
"When a task goes smoothly, your brain remembers" (Score:2)
Yeah like we all learn from our successes, rather than our failures.
Unless they probably mean a dopamine hit, which is also not a good indicator of something being a good idea for using.
This story and this summary reads like a hallucination party - no thanks.
Assembly (Score:2)
Then why not tell the AI to write the code in highly optimized Assembly?
Re: AI (Score:3)
AI will touch a line of code of mine only after I'm dead, buried and eaten by a whole bunch of macro/micro-organisms far more worthwhile than AI will ever be to me. I might be getting a little long in the tooth, but predictive neural networks can't and won't surpass the human brain at programming for a long while yet.
It did not (Score:3)
> AI absorbed the friction that made shell scripting painful.
What friction? What pain? Shell scripting is easy, and if you don't understand or remember how to do something, you can trivially find the answer with google unless you are truly completely incompetent at determining which search terms to use. Or often, find the answer by studying scripts which are already on your system! Don't be fucking helpless, which is what AI will make you by doing it for you.
Making things easier for package maintainers was one of the big excuses for adopting systemd. Now we have a whole new class of problem, problems with systemd. Meanwhile shell scripting is a core Unix[like] feature. If you don't understand it well enough to do init scripts, you aren't qualified to be a packager either.
Yes. With me it is. (Score:2)
I feel totally confident solving problems with PLs I wouldn't have touched with a ten-foot pole just a year back, due to AI.
Example: The legacy application I am currently maintaining and replacing is totally borked with piles of spagetti-code and shitty, amateurish or simply non-existent architecture. However, I do have to add logic to this already unmaintainable system so I often just push larger portions of that logic further down into the DB and SQL.
SQL _is_ turing complete, but actually developing appli
Re: (Score:2)
Interesting post--I have also found that LLMs are extremely good at creating and optimizing queries and, as you mention, PL.
I also am dealing with some legacy software. Many years ago I wrote a text file to SQL updater that keeps an SQL database in sync with the legacy application text file databases. With LLM I have created and moved some annoying chunks of logic into triggers, a couple of GENERATED columns, etc. I am reasonably good at SQL. I always spent way more time working on triggers than I probably
The one bright side of LLMs (Score:2)
....is that people will hopefully use less Python and stop using those bullshit "It makes me more productive excuses."....when they meant to say "I don't know how to use any other language". If you code for fun?...play with whatever toy you like...when you do this professionally, keep scripting languages out of programs that are meant to scale and do real work. A fuckton of Python has to be rewritten every goddamn year. Some student or intern writes a cool app...it gets users and then it has to be rewritt
Re: (Score:3)
> scripting is not coding
So, what is it then? Spelunking?
Re: (Score:2)
scripting? But yeah, while I can see how one could draw distinctions between 'programming' and 'scripting', I've also seen people argue that C/C++ are not 'programming' either, so the 'this isn't programming!' always seems a bit silly.
Re: (Score:2)
Programming isn't coding. Programming is the activity of making programs: coding is expressing data in the form expected by a process. While it's true that programs are data, coding is a pretty small part of programming and a much larger part of certain data entry jobs. (To take an extreme example, there are jobs advertised which primarily consist of coding medical records in HL7).
Re: (Score:2)
Eh, I am old enough to remember the arguments about if writing things in high level languages counted as 'programming' since you were no longer making the program that goes on the hardware.
Re: cool and all but.... (Score:2)
Advanced config file writing
Re: (Score:2)
> Advanced config file writing
This is a fundamental misunderstanding of Unix.
Re: (Score:2)
Bash scripting is coding where the commands you are running are your external functions from a (maybe 3rd-party) API package.
You are running such a function, taking the output of it, manipulating it (likely with another external function), and using that as input for another function.
It is exactly, in all regards, a coding language.
It even has flow control (if, elif, else), loops (while, for, etc), and all the other stuff you expect of a coding language.
You'd prefer C with security vulnerabilities? (Score:2)
> scripting is not coding. why isnt C and C++ usage increasing rapidly now that its supposedly "easy" ? How but churning everything out in Assembler ? just tell the AI what to do. Why is everything a crappy script with giant security holes ? oh wait. its because AI cant actually code. just regurgitate out trivial B.S. which interns used to do.
Baby-steps...while TypeScript has dogshit performance, at least it has compiler support and is easier to maintain than JavaScript or Python. You tell someone who barely learned Python to go write C and they'll not understand memory management and security very well. But hey...who has ever written insecure code in C? As a daily Claude user...it can't reliably write Java, the most popular business open source language, that compiles reliably. I think it compiles about 60, tops 75% now. LLMs cannot be use
token efficiency (Score:2)
One of the issues with C and C++ and many other "verbose" and overly "ceremonial" languages is that they are very token inefficient.
Expect languages that get more done with less code to be more popular going forward. The models have limited context and you don't want to waste half of it on boilerplate.
Re: (Score:2)
because aside from the core GNU/Linux stuff, there's not a lot of C++ out there compared to TS/JS and shell (or even Python and perl and php).
C++ coding is rarely just 'raw'. usually the bulk of what you're doing is integrating multiple libraries together. How do you talk to your database? How do you create an API and/or a webpage and/or a UI? Each of these decisions requires integration that isn't published as well as the thousands of examples out there on stackoverflow for an AI to gobble up.
So internally