How Effectively Agents Write D

[bachmeier]

On Friday, 6 November 2026 at 15:23:33 UTC, Peter Alexander wrote:

> Keen to hear any and all feedback. Also interesting in what 
> other ideas people have for how to measure D for agents.

This is just my opinion, may cause some people to explode. Claude 
gets all the attention. My opinion is that D should stand out by 
working with cost-effective options (GLM 5.3 Flash, Deepseek 4.1, 
and those types of models) and work with open agents like pi.dev.

I recently ported a not insignificant C codebase to D using 
Claude. It worked, but I did not like that I had to use their 
proprietary harness, who knows how much it will cost after the 
IPO (I don't know what their motivation would be to continue 
offering subsidized plans at that point), and it is ridiculously 
slow. Here's the repo: 
https://github.com/bachmeil/dgretlpreview/tree/main

The speed difference vs Gemini, which writes D very well, made 
the experience semi-miserable. Even a port of a 100-line C file 
would take six minutes or more with Claude. I tried Opus and it 
was too clever by half and I even found some bugs, so went back 
to Sonnet real fast. I also use GPT 6 Luna, and it's fast, cheap, 
and knows a lot about D.

On a different note, what I would find interesting is performance 
of different models porting Rust projects like ripgrep into a 
sensible language like D. Some of us still like to read code.