Since the beginning of my career where I worked at Olivetti research and was part of a team (AI team) that developed a declarative programming environment using natural language and second-order predicate logic, I have followed developments in this field with great interest, and as the technology has evolved I have systematically tested its ability to generate code (X++) for Microsoft Dynamics 365 F&O.

But let me show you some snippets of some of my recent experiments. Note that what I'm showing you here was pretty representative of the experience across the board, even though I was left with the feeling that ChatGPT was a little better.

Prompt > how do I incorporate the cost center data in the CustAgingReportTmp temporary table? 

I will save you the full response, and show you just some small samples of the code generated and the interactions.

The following was systematic and representative of the back and forth this stuff was.

I'm not sure this would be good guidance and help for a junior developer without much knowledge of Microsoft Dynamics 365 F&O. My experience is that working under the supervision of an experienced software developer would be better. The code is on the surface flawless, but wrong except for trivial stuff.

The most dangerous to me is that it looks really sensible to an inexperienced developer, and given my results there is a serious risk that some stuff of the generated code would actually compile but be completely inappropriate and end up in a product.

It is of course not the control system of a life critical system, but my conclusion is that the ability of AI tools (and an LLM does not fit my definition of AI, but never mind), is at least when it comes to generating any serious programming code, very questionable and more trouble than it's worth for now.