www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - AI Experience with D

reply harakim <harakim gmail.com> writes:
I finally got a Claude account for myself and decided to rewrite 
a program to create USB drives for my car stereo. My old one was 
command line driven and was hard to use.

I was hesitant to use D because there is more documentation for 

going to involve UI libraries, which are rarer still.

I used Gemini to work through some of the specs (3ish minutes) 
and came up with a plan to use GTK-D (it also suggested raylib 
and dlangui). I built a 3 pane editor for the file explorer, the 
albums pane and a playlist pane. This is due to how my car stereo 
reads USBs and I won't get into it here.

My experience: Claude had no trouble with D. This was vastly 

python scripts it generates often have issues though it can 
usually resolve them without intervention. For D, every time it 
was done, I ran the program and it worked, with only one 
exception. To be clear, this is significantly better than my 
experience at work (2 different companies, both on claude).

In the entire time I was developing this program, there was only 
one issue running and it was with GTK-D. Claude assumed that some 
iterator would be initialized but it was not. All I had to do was 
dump the program output from bash back into Claude and it looked 
at the source in the dub folder and fixed it.

The 3 pane UI took just over 20 dollars but did take about an 
hour wall-time while I was doing other things like folding socks. 
Generating the json export file and outputting is very simple so 
I'm expecting about 25 dollars and a little over an hour for a 
complete application.

The only time my developer skills were necessary were when I 
asked Gemini which UI library to use (I know UI libraries exist) 
and then knowing enough to paste the bash output of the app with 
the error into claude.

So I will have a completely functional app to copy music to my 
car for about 25 dollars and I didn't even need to understand 
code to do it. This was significantly better than what I get with 


I am an infrequent visitor this days, but I wanted to share this 
as a bit of good news.
May 23
next sibling parent "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
If you haven't already try GiD as it is more up to date than Gtk-D.

IDE's like IntelliJ will also provide context such as error messages 
automatically. Kilo plugin can do this, although I don't really use it.

The interactive nature of it is pretty awesome.
May 23
prev sibling next sibling parent reply Anton Pastukhov <mail anton9.com> writes:
On Sunday, 24 May 2026 at 05:54:22 UTC, harakim wrote:


 My experience: Claude had no trouble with D.


Yup, LLMs are comparatively good with D, to my surprise. It seems 
there is enough training data around for them to make sense of 
it. I'm wondering how newer languages (Nim, Zig etc) fare.
May 24
parent Paolo Invernizzi <paolo.invernizzi gmail.com> writes:
On Sunday, 24 May 2026 at 08:19:55 UTC, Anton Pastukhov wrote:

 On Sunday, 24 May 2026 at 05:54:22 UTC, harakim wrote:


 My experience: Claude had no trouble with D.


 Yup, LLMs are comparatively good with D, to my surprise. It 
 seems there is enough training data around for them to make 
 sense of it. I'm wondering how newer languages (Nim, Zig etc) 
 fare.


They are really good with languages that have the luxury of 
turning lots of RT errors in CT errors.

D is among them, claude-code opus 4.7 is _really good_ in 
handling D codebase.

/P
May 24
prev sibling next sibling parent reply Walter Bright <newshound2 digitalmars.com> writes:
On 5/23/2026 10:54 PM, harakim wrote:

 I am an infrequent visitor this days, but I wanted to share this as a bit of 
 good news.


Thank you for taking the time to post this! It's perfect for the n.g.!

May I suggest making a HackerNews post out of this? Or I can do it for you?
May 25
parent harakim <harakim gmail.com> writes:
On Monday, 25 May 2026 at 17:35:39 UTC, Walter Bright wrote:

 On 5/23/2026 10:54 PM, harakim wrote:

 I am an infrequent visitor this days, but I wanted to share 
 this as a bit of good news.


 Thank you for taking the time to post this! It's perfect for 
 the n.g.!

 May I suggest making a HackerNews post out of this? Or I can do 
 it for you?


I'm not good at making promotional posts or I would have done a 
better job at my message above, but you are free to use my above 
post or parts of it in any way you'd like.
May 27
prev sibling next sibling parent Mark Davies <savageshrimp pm.me> writes:
On Sunday, 24 May 2026 at 05:54:22 UTC, harakim wrote:

 I finally got a Claude account for myself and decided to 
 rewrite a program to create USB drives for my car stereo. My 
 old one was command line driven and was hard to use.

 [...]


One of the good things about D with AI is that you can point it 
to the import directory of the compiler and have it look up the 
source and unit tests to see how it should be using a particular 
module.
May 25
prev sibling parent reply =?UTF-8?Q?Ali_=C3=87ehreli?= <acehreli yahoo.com> writes:
On 5/23/26 10:54 PM, harakim wrote:


 for about 25 dollars


I heard from a friend that AI coding is currently highly subsidized. 
Does anyone know what the actual cost is today or will creep up to in 
the future?

Ali
May 28
next sibling parent reply monkyyy <crazymonkyyy gmail.com> writes:
On Thursday, 28 May 2026 at 18:11:32 UTC, Ali Çehreli wrote:

 On 5/23/26 10:54 PM, harakim wrote:


 for about 25 dollars


 I heard from a friend that AI coding is currently highly 
 subsidized. Does anyone know what the actual cost is today or 
 will creep up to in the future?

 Ali


open weight models on 3rd party hosts likely are at cost, you 
could try qwen-cli on whatever chinese host it suggests(they 
stopped giving away compute) and see whats the real cost/benefit 
for the future
May 28
parent reply Sergey <kornburn yandex.ru> writes:
On Thursday, 28 May 2026 at 18:31:11 UTC, monkyyy wrote:

 cost/benefit for the future


I can’t believe I’m saying it, but monkyyy is right
May 28
parent monkyyy <crazymonkyyy gmail.com> writes:
On Thursday, 28 May 2026 at 19:33:40 UTC, Sergey wrote:

 On Thursday, 28 May 2026 at 18:31:11 UTC, monkyyy wrote:

 cost/benefit for the future


 I can’t believe I’m saying it, but monkyyy is right


???

when was I ever wrong
May 28
prev sibling parent Lance Bachmeier <no spam.net> writes:
On Thursday, 28 May 2026 at 18:11:32 UTC, Ali Çehreli wrote:


 will creep up to in the future?


Well, this is the real question, isn't it? There's no way anyone 
can predict this. Hardware costs, electricity costs, potential 
data center supply issues, how often models are trained and how 
expensive it is, etc. will be determinants. There's no way to 
forecast this. Zuckerberg has spent a lot developing his own 
models so he's not caught by large price increases.

I'm able to run local models at home and at work, so I'm in a 
good spot as far as that goes.
May 28