What languages do you code in?

[NoPeace]

I code in English. I just type it in Claude and Claude does the rest.

Problem with that is that it doesn't 100% work. You'll always need to go back in and fix things. It gets it 90-95% of the way and then you need to do the cleanup. Not to mention, Claude does love to over complicate things and can make run times slower because of it.

 

[auglyn]

Problem with that is that it doesn't 100% work. You'll always need to go back in and fix things. It gets it 90-95% of the way and then you need to do the cleanup. Not to mention, Claude does love to over complicate things and can make run times slower because of it.

I could always switch to ChatGPT

 

[NoPeace]

I could always switch to ChatGPT

I find Claude does a much better job at writing code. ChatGPT is great for simple coding things, but on more complex tasks Claude does a much better job. ChatGPT likes to over complicate things and it takes a lot longer to figure out WTF it's doing when trying to troubleshoot.

 

[auglyn]

I find Claude does a much better job at writing code. ChatGPT is great for simple coding things, but on more complex tasks Claude does a much better job. ChatGPT likes to over complicate things and it takes a lot longer to figure out WTF it's doing when trying to troubleshoot.

I could always switch to Google Gemini 4 Argon.

 

[kmarkie]

I have found C23 to be a vast improvement over the over the C99 projects that i have worked on for close to a decade. The same thing could be said about Java 6, 7 and 8 that many legacy projects unhappily enough still freaking use. Most at least use Java 8 but once you start using anything over 21 its getting pretty great.

Virtual Threads are one of my most loved and most hated features, normally love but gawd programming is a labor of love people forget that large LLM's are just generating existing source code it has been trained on so if you want something unique and production quality. its harder.

It's hit or miss even with a tightly trained LLM focused on C for example focused using NASA's JPL the power of ten and the more recent JPL 1.0 pairing select features with MIRSA among others it can be much nicer.

This still requires a programmers mindset along side architectural design per product this is where LLM's tend to fail if you are unsure of how something works then how do you know what to fix?

It will gripe about Use after free and other weird issues that once manually tested are false positives so LLMs of all kinds can be great if used properly as a tool but as a full replacement no.

The logic behind this is when something does break from purely LLM Generated code they call me because they do not understand even the basics in most cases.

This is my professional experiences .