Test Driving Generative AI in Oracle Apex

I have started to experiment with the onboard AI Assistant that is part of Apex 24.1.  It is remarkable, how much progress there has been in the AI world since ChatGPT was released.  What I saw at Cloudworld this year was a very pragmatic way to utilize external Large Language Models.  What I see in Apex now is a way to both use them within development and build AI capabilities very naturally within other applications.  

There is a true conversational user interface to App Developer.    You can describe the application that you are trying to develop and have the assistant describe back the application in terms of the pages that it will create.  

You need to already have defined some of your data model to utilize this capability.  I found that the assistant is heavily influenced by the table definition.  This is not surprising as Apex has been very good at making a guess at your application design from the way you are storing the data for some time.  I know this is Oracle's first step into this field, but It might be very useful to ask the user to describe their application and use the parts of speech to make an attempt at the data model.  This is what we do as analysts and designers.   

The assistant knows the objects that are within the workspace so you can make more natural language requests and have the assistant make its best guess at the SQL you were trying to construct.  

For myself I find that that I start at Query Builder.  I found that the time spent refining my prompts might be as productively used in Query Builder.    After spending a lot of time in my life constructing SQL statements, I find that my brain makes a good attempt at them by itself, but working from a data model diagram makes it easier. 

The one area that I tend to spend far to long in query development is getting outer joins, left joins, right joins correct.  I have to admit that the syntax feels much father away from standard English.  The assistant seems to make this much easier.  

Creating  PL/SQL also seems much accelerated in asking the assistant to prepare the code.   

I look forward to being able to use the "Explainer" and "Improver" aspects of the assistant.  It looks as if they will remove the frustration of being stuck on how a piece of code is working, finding 50 dead ends before finding some simple solution.  

It is interesting that the assistant has great knowledge of SQL and PL/SQL, but although javascript is incorporated into SQL Workshop it is not yet incorporated into the assistant.