AI Redux

A lot of the comments on my prior article had to do with me not asking the right question about the AI engine. There were no comments about how, who, or why the AI engines were trained on what code. Well, is that the problem? Must my question be specific enough that the AI engine (ChatGPT, Llama, or Gemini) does not interpret my question or request correctly?

Obviously, the AI engine has not been in front of clients when they want a change the current code or have this great idea of what this new application should do. I cannot tell how many meetings I have been to where the clients tell me what they want. Then I spend a week or two coding and show them the result. The clients then say that is not what they said even though they signed off the build document.

After an hour or so of heated discussion, I have to redesign the build document. The clients have to sign off on the revised document but as has been the case most of the time, the clients will not pay for the added cost of the redesign. This process for several clients went on for years until they wondered why the bill was so high, duh!

Moreover, in all of the articles I have read, AI is going to be my little helper or write all of the code for the application. Unless the clients are an active part in the design and see the result within a few minutes of their request I cannot see this working in the next 5 years or so.

U2logic has an XLr8Developer that allows you to create a form with or without objects so that the clients can see where their prompts are going to be. In the design meeting, I can move the prompts around without code and just be doing something similar to copy and paste. It takes me about 5 minutes to develop a form template that can be manipulated to move the prompts around to match the client's expectations and needs.

AI code generation is just a minuscule process in the creation of an application such as:

• What database should be used
• What user interface are we using browser, telnet, or windows
• What is the cost of database licenses
• What is the volume of transactions that this application is going to produce
• Do the transactions need to be recoverable
• What computer system is the software going to reside on

Finally, someone asked what UniBasic code I would produce that would answer the AI question I posed: "Create an MVBasic program that reads, writes, and deletes from a file. Please optimize the code." This code was written and tested on Universe 14.2 Beta.

   JSON = \{"Id": 178, "Record":["Bob's Pan Shop","1234 East Easy Street"]}\ 
   LOOP 
        CRT 'Enter method ': ; INPUT METHOD
        IF METHOD = 'Stop' THEN EXIT
        GOSUB PROCESS.METHOD
    REPEAT
     STOP
PROCESS.METHOD:*
     OPEN '','CRM.CUSTOMERS' TO CUSTOMERS THEN 
        obj = JSON->$fromjson()
        ID = obj->Id
        BEGIN CASE
           CASE METHOD = 'Read'   ; GOSUB READ.RECORD
           CASE METHOD = 'Write'  ; GOSUB WRITE.RECORD    
           CASE METHOD = 'Delete' ; GOSUB DELETE.RECORD
           CASE 1 ; RETURN.MSG = '~Invalid method'
        END CASE
     END ELSE RETURN.MSG =  '~Customers failed to open'
     CRT RETURN.MSG
     RETURN
READ.RECORD:*
     READ RECORD FROM CUSTOMERS, ID THEN
        RETURN.MSG = RECORD
     END ELSE RETURN.MSG = '~Reading failed'
     RETURN
WRITE.RECORD:*
     NEW.RECORD = ""
     NEW.RECORD<1> = obj->Record[0]
     NEW.RECORD<2> = obj->Record[1]
     WRITE NEW.RECORD ON CUSTOMERS,ID
     RETURN.MSG = 'Record written'
     RETURN   
DELETE.RECORD:*
     DELETE CUSTOMERS,ID
     RETURN.MSG = 'Record deleted'
     RETURN
  END