AI - Revisited in 2026

I have tried and used most of the LLM AI tools. I found them to be helpful but I would not use them for an entire application. I asked an AI engine to analyze a working program that was using UDO. Here is my favorite comment:

You caught me red-handed—that's two swings and two misses on syntax! MAPTOJSON isn't a native UniData function either.

In Rocket UniData, JSON handling is split between specialized API extensions and built-in system subroutines, depending on the version.

Is this something you really want real user to be using?

​@Doug Averch, I would say MAPTOJSON wouldn't be native.  Have you tried uploading the documentation that use MAPTOJSON?

I use AI as my senior programmer which use to be StackOverflow for years and years. Although there was very little UniBasic on StackOverflow, I did find it useful when I had questions on UniObjects for Java, Java Script, Java, HTML, and CSS for examples. Rocket Software announced that they would be integrating AI into Universe/Unidata so I thought lets see what the AI systems can do.

Me: What does this section of code do? (UniBasic Universe Code not shown here)

This routine generates an accounts payable ledger for a vendor, returning paired Invoice-to-Check payment history along with multi-currency and reconciliation details, while fixing broken cross-references between invoices and checks in the process.

Me: Would there be a better way to write this in Universe UniBasic?

Array Building Syntax

Instead of building delimited strings like JSON.NAME = "..." and converting commas via CONVERT ',' TO @FM, standard dynamic array assignment syntax JSON.NAME<1> = "InvoiceNo", JSON.NAME<2> = ... or field insertion (INS) is cleaner and less prone to delimiter collisions.

Me: How can I use UniVerse built-in JSON creation functions instead of manually building dynamic arrays?

Starting in UniVerse 11.3.1+, Rocket UniVerse introduced native JSON creation and generation APIs, replacing manual string concatenations and dynamic matrix structures like JSON.VALUE<1, JSON.CNT> with clean object transformations.

Key Native UniVerse JSON Functions

UniVerse provides two main approaches to generate JSON natively:

1. MAPTOJSON / JSONToMap (UniVerse Object API): Uses key-value maps (MAP data type) to create real JSON objects or arrays of objects.

2. JSON.XLS / RECORD.TO.JSON Subroutines: Converts dynamic arrays or record structures directly into JSON strings.

Me: Wow, this MAP and MAPTOJSON is not in Rocket Software Universe 14.2 or lower.

You are correct to double-check that—MAP and MAPTOJSON are actually features introduced in UniData (via Python integration / Object Basic) and Rocket D3, or via Python bindings in newer UniVerse releases, but standard UniVerse Basic (11.x through 14.x) does not have native MAP() or MAPTOJSON() functions built into UniBasic syntax.

Hence, the AI engine gave me functions that I cannot find in the Universe UniBasic manual. This is even worse than a rookie programmer because at least the rookie would know that their code does not compile. But the AI engine does not care because it is spitting out things that may or may not be correct to answer your questions.

And you must ask the right question to to get the write answer or you will be writing applications that do not scale or worse they do not compile. Asking AI to create an application in any language without defined specifications from stake holders is a guarantee that the code will be not usable, not scalable, and not maintainable.