AI-Assisted Coding and Prompt Engineering

 
A place to discuss practical AI-assisted coding practices and experiences.
 

"It's just a tool"


 

https://www.mql5.com/en/forum/504817

Personally, I was impressed by the fact that ~11,000 lines of MQL code created with AI assistance compile without errors. That is to say, I was impressed by the length of the source code. Although, to be fair, I should note that I didn't run that code, and I don't know what percentage of it was generated by the AI.

Resolving Conflicting Bias in GOLD Confirmation Logic (MQL5)
Resolving Conflicting Bias in GOLD Confirmation Logic (MQL5)
  • 2026.02.06
  • www.mql5.com
Hi all, I’ve been experimenting with a structure-based EA approach on XAUUSD where internal market behavior is evaluated together with a few correl...
 
Vladislav Boyko #:

https://www.mql5.com/en/forum/504817

Personally, I was impressed by the fact that ~11,000 lines of MQL code created with AI assistance compile without errors. That is to say, I was impressed by the length of the source code. Although, to be fair, I should note that I didn't run that code, and I don't know what percentage of it was generated by the AI.

@GoldRankers, why is all the code in a single file? Is this more convenient for you or is it more convenient for the AI?
 
Vladislav Boyko #:
@GoldRankers, why is all the code in a single file? Is this more convenient for you or is it more convenient for the AI?
In AI-assisted workflows, preserving full context is critical for consistent outputs. Keeping the project in a single file helps maintain continuity when working with tools like Claude.ai via the web interface.
However, I’m fully comfortable restructuring the codebase into a modular multi-file architecture — it’s technically straightforward and can be implemented at any stage. 
 
GoldRankers #:
In AI-assisted workflows, preserving full context is critical for consistent outputs. Keeping the project in a single file helps maintain continuity when working with tools like Claude.ai via the web interface.
However, I’m fully comfortable restructuring the codebase into a modular multi-file architecture — it’s technically straightforward and can be implemented at any stage. 
Does the AI ​​edit that huge file and provide you with a new version to download? Or does the AI ​​provide you with a code snippet, and you integrate it into the file manually?
 
Vladislav Boyko #:
Does the AI ​​edit that huge file and provide you with a new version to download? Or does the AI ​​provide you with a code snippet, and you integrate it into the file manually?
YES, It's edit the huge file and push it into my github 👌
I do understand your concerns, please try Claude.ai , you will be amazed 
 
GoldRankers #:
I do understand your concerns

I don't know what you mean by concerns, but in any case, I have no concerns.

GoldRankers #:
YES, It's edit the huge file and push it into my github 👌

Got it. I haven’t tried Claude, but I tested OpenAI Codex, and it also pushed changes to my GitHub repo.

For now, I don’t actually need generated code, so I found Codex less convenient than the regular web chat (ChatGPT 5.2 Thinking), which I use a lot for refactoring code I’ve written. More specifically, in most cases ChatGPT helps me choose identifiers and write commit messages. I should note that it handles those tasks really well.

[Deleted]  
GoldRankers #:
I did, ❤️
Which AI it is? ChatGTP, Grok, Gemini, or other?
 
Vladislav Boyko #:
Does the AI ​​edit that huge file and provide you with a new version to download? Or does the AI ​​provide you with a code snippet, and you integrate it into the file manually?
Well , I add more functions to the EA , now it rach 12K line and it works 100% :) I am sure 5 years from now , we have to start finding new job LOL
 
Shanmugi #:
Which AI it is? ChatGTP, Grok, Gemini, or other?
https://claude.ai