IBM Developing AI Coding Assistant for IBM i

May 22, 2024 Alex Woodie

During his keynote presentation at the POWERUp 2024 conference Monday, IBM i CTO Steve Will announced that IBM is actively exploring ways to incorporate generative AI into the IBM i platform. The most promising of three individual but related projects is a plan to build a large language model (LLM)-powered coding assistant for IBM i that will initially provide three functions to support RPG development, with more expected in the future.

The genesis of the coding assisting project for RPG, which doesn’t have a name yet, can be traced to last year’s launch of watsonx Code Assistant for Z, which converts COBOL to Java, among other capabilities. After the mainframers got a copilot, members of the IBM i midrange community pestered IBM about whether IBM i would be getting something, too, and if so, then when?

“Some of you in the audience contacted us and said, Is there going to be some code assist for IBM i and is it going to transform RPG into Java?” Will said during his keynote at Fort Worth Convention Center May 20. “And we were pretty darn sure that that was not the right thing to do for this customer base. But we began thinking about what we might want to do in this space.”

After convening with a group of advisors and conferring with various groups’ councils, Will and his team established what direction they would take.

“What we decided that we really needed to do was to create a code assistant that could help programmers’ existing RPG, not move from RPG to something else,” said Will, who also holds the title of distinguished engineer and chief architect for IBM i. “Our biggest problem is that when you hire a new programmer, it can be difficult for them to take this massive amount of RPG that exists in the world today and add something to it, keep it up to date, etcetera.”

IBM i CTO Steve Will delivered a keynote address at POWERUp 2024 on Monday May 20, 2024.

That will be the first function delivered by the new coding assistant that IBM has planned for RPG: explaining what a section of RPG code does. LLMs have proven themselves to be as adept at understanding and generating computer languages as they are natural languages, such as English or French. So summarizing existing RPG code will be job number one, or numero un.

Considering the amount of old, poorly documented RPG code floating around, this GenAI-powered language understanding capability would be well received, Will predicted.

“This is one of the issues we tend to find with new programmers going into old code,” he said. “It doesn’t matter how well somebody thinks they documented it — if it was written in RPG 2, it looks really hard to understand to a brand new programmer. And so if you could have a tool that would do that, that would be very, very beneficial.”

Once an LLM “understand” a programming language like RPG, it can start writing it. So that is job number two for the forthcoming IBM i copilot: generating RPG.

“If you’ve used any sort of generative AI at all, one of the things that you know that you can do is you describe what sort of software you’d like, what sort of subroutine you’d like it to write, and it will write it in that language,” Will said.

“Now, it might not be the best performing version of it, but it’ll probably get pretty close,” he continued. “And if you were a non-RPG programmer and you said ‘Please learn, write me something that does this’ and the RPG came out, that would be a very beneficial thing to you. You could start there and then figure out what else needs to be done.”

Will shared this slide on the RPG code explanation capability of the AI code assistant POC.

The third function that IBM is targeting is the writing of test programs.

“Wouldn’t it be cool if you could give this tool a program and then say ‘Write me a test program that will test all the paths or the edge cases or whatever for this?’ Wouldn’t that be cool?” Will said. “And that became one of the core things that our project would want to be able to do.”

This project is very new, and there are still a lot of unknowns around it. But Will said he’s already seen a proof of concept (POC) for a GenAI-powered AI assistant (some call them copilots) that was trained on RPG. The POC was developed by an IBM i customer that he didn’t name.

“They worked on it for a few months and they produced a passable result,” Will said. “I mean, it’s not good enough for prime time, but it proved that you could do all of those things.”

Will shared some screenshots from the POC that showed the AI assistant generating free format ILE RPG procedure based on a table defined in SQL. Another screenshot showed the AI assistant being presented with “a bunch of gobbledygook stuff that really was RPG” and then explaining what the RPG did in plain English.

“And then we were able to say, OK, given the procedure I’m about to give you below and the DDL associated with it, write me a test program that covers the edge cases,” Will said. “And sure enough, it started writing a program for each test case…”

“That’s pretty cool, I would say. Sound cool?” he continued. “If you had a tool like that that you could make available to all of your developers, particularly the ones who haven’t had any experience with RPG, wouldn’t that be a really great thing to have?”

While IBM is targeting those three functions explained above, it’s already entertaining other ideas. For instance, the assistant could be used to migrate old System 36 code to free format ILE RPG IV code, Will said. Or the code assistant could suggest ways to replace an old greenscreen interface with newer open source technology.

“Or what if we could transform the old data definitions that we had into modern SQL and Db2? Wouldn’t that be cool?” Will asked the audience. “Yes, that would be cool. We could do so many things. That’s the power of generative AI, is that if you train the model, it can help you do that.”

It’s unclear what the timeline for the delivery of the AI coding assistant is. We do know that Will and his team intend to use IBM’s own Watsonx LLMs for this project, which makes sense. But that brings up another topic: IBM needs RPG.

IBM created the RPG language, and it owns it. It has a fair amount of RPG that it has developed over the years for various demos and what not. But that’s still not enough code to train an LLM on. That’s why Will is soliciting RPG from the IBM i user base.

Will shared this slide on the three AI use cases IBM is working on for IBM i.

“Any of you who would like to work with us to submit code that can be used to train, we will work with you now,” he said. “I can tell you for sure that anybody who wants to donate code and have it in the open, we’re going to accept it, period.”

IBM isn’t quite ready for the IBM i community to start sending it RPG. There are some details that it hasn’t solidified yet, Will said. Folks interested in donating their RPG will have to sign away the rights to it to allow IBM to use it to train its new LLM, but the actual process for that hasn’t yet been created.

During his keynote, Will discussed two other AI initiatives IBM is working on with IBM i. The first one involves using AI against IBM i transactions.

“There are AI workloads that can sit next to the transaction stream and help you notice things, help you figure out trends, help you figure out the fraud may be happening, etcetera,” Will said. “What you’d like to be able to do is to have that AI workload come up very easily and connect back to that data and transaction source.”

IBM needs two things to happen to make this possible. The first is open source technology to connect the transactions in IBM i to the AI. “There is a set of standards, a set of protocols, that the whole industry is gathering around and we need to make sure that that’s available,” he said. “We’re investing in that right now. This is a directional thing. It’s happening behind the scenes. And in the future you will see us talk about connecting IBM i transactions more and more to AI because we’ve invested in those technologies.”

The second additional AI project IBM is working on involves operational use cases for using AI on internal data generated by IBM i.

“If we can figure out how to get the right transaction data, like messages that go to QSYSOPR and so on over to an AI workload that can do automated operations, it could then help you run your machine,” Will said. “And so that’s another thing that we’re focusing on–what are the necessary open technologies that carry information about operations and networking and performance to an AI workload so that it can help you do things. That’s another directional thing that we’re headed towards.”

RELATED STORIES

The Time Is Now To Get A GenAI Strategy

Top Priorities in 2024: Security and AI

Thoroughly Modern: Proceed With Caution With AI In The Landscape Of Cybersecurity

IBM i Shops Are Still Getting Their Generative AI Acts Together

IBM To Add Generative AI To QRadar

How Long Before Big Blue Brings Code Assist To IBM i?

Generative AI Is Part Of Application Modernization Now

Sticking To The Backroads On This Journey

With Fresche’s New CEO, There Are No Problems, Just Solutions

Enterprises Are Not Going To Miss The Fourth Wave Of AI (The Next Platform)

IBM Introduces watsonx For Governed Analytics, AI

Technology Always Replaces People While Augmenting Others

Generative AI: Coming to an ERP Near You

It Is Time To Have A Group Chat About AI