Beta For RPG Coding Assistant On Track for 2Q25
January 20, 2025 Alex Woodie
IBM is ramping up development of the AI-based RPG Code Assist and expects to have a working prototype ready for testing by the end of the first quarter, IBM i chief technology officer Steve Will said in a webinar last month. The goal is to have a beta ready for during the second quarter and general availability hopefully in the second half of the year, he added.
Large language models (LLMs) such as OpenAI’s GPT and Google’s Gemini have taken the world by storm, and companies are investing tens of billions of dollars in the nascent tech to gain a competitive edge, or simply to keep up with competitors. Just about any job where the primary interface is a natural language could potentially be impacted by advances in LLMs. That means that customer service workers could be replaced (or augmented) with sophisticated chatbots, while software engineers are boosting their productivity with coding co-pilots. Even some journalism jobs are being replaced by robot writers (ahem).
The RPG Code Assist falls into the popular coding co-pilot category of generative AI (GenAI) tools. Microsoft owns the first-mover advantage here with GitHub Copilot, which it launched in October 2021, more than a year before OpenAI dropped ChatGPT onto the world in late November 2022.
The Code Assist for RPG that Will unveiled during the COMMON POWERUp conference last May in Fort Worth, Texas, is based loosely on the Code Assistant for Z that IBM launched in October 2023. That offering uses IBM’s in-house Granite AI models to accomplish its primary function, which is converting COBOL code into Java.
Rochester’s co-pilot will be similar in some ways, but different in others. While it will also use IBM’s Granite foundation AI models (among potentially others), Will has made it clear that there will be no RPG-to-Java conversions, and that there are already decent, non-AI code converters out there.
IBM has experienced ups and downs with the development of its Code Assist for RPG product. (Image courtesy IBM)
Instead, there are three main goals with the Code Assist for RPG, including code explanation, code generation, and test case generation, all for RPG. “We want to create a tool that will help programmers work with existing RPG,” Will said during a December IBM i Guided Tour on the coding assistant.
The number one priority for the Code Assist for RPG team is to create the RPG explanation function, which Will repeatedly called Explain. The team is currently focused on delivering that capability with the alpha release of the new tool, which is currently on track to occur before April 1. Depending on how quickly the other two capabilities – code generation and test case generation – come around, they may or may not be included with the beta IBM wants to release before July 1, or even the GA of the final product, which Will says should be in the second half of 2025.
“I’m pretty confident we can get an Explain out there with a product around it in 2025, in the second half,” Will said. “That’s what we’re shooting for.”
Some aspects of the new tool have been nailed down. For starters, it will be delivered as a plug-in for VS Code, the Web-based IDE that’s become very popular among IBM i coders. IBM may deliver an Eclipse version of the coding assistant, thereby enabling Rational Developer for i (RDi) users to benefit from it, but not on the first go-around.
“We’ve got to develop with something first,” Will said. “VS Code is not only more nimble to work with, its got a lot more support in the open source AI world within IBM and external to it, so we are going to have to look at this [RDi support] as something that we would have to do later and prioritize it in with other things.”
Will also announced that the coding assistant will live on the IBM Cloud to start out. IBM i customers are requesting a version of the GenAI product that lives on-prem, but that is not a priority at this point, Will said. “We know that’s a requirement,” he said. “But at the first instantiation of this, we’re not sure we can make an on prem version of it at the beginning.”
The Code Assist for RPG will support English out of the gate. The question is whether or not it will support other languages, too. Will said that he originally was resigned to targeting just English at the start. However, he said that he recently learned that IBM may actually be able to support additional languages earlier than he initially thought, which could be good news for some of Rochester’s Japanese clients.
Rochester is building Code Assist for RPG on IBM Granite models. (Image courtesy IBM)
Developing GenAI tools around LLMs is a completely new function for Will’s team of engineers in Rochester and other IBM labs, and that’s taken some getting used to.
“Let me just tell you guys, creating and training a large language model, creating a code assistant around that, is nothing like creating an operating system,” said Will, who also holds the titles of IBM i architect and Distinguished Engineer. “So we have had to learn a lot of stuff to be able to do this. It’s been super exciting. I have learned more in the last nine months than I’ve learned in the last several years, it feels like, because we’ve had to learn so much and it’s been exciting.”
Will and his engineers have leaned heavily on the IBM Z team that developed the Code Assistant for Z assistance. He’s also sought the advice of AI experts outside of IBM who have more familiarity with LLMs and AI. That has helped the Rochester team ride the highs and lows associated with AI development.
“We have talked to a number of people who have trained large language models, and what they’ve told us is that after we start training the model with more code . . . what we’re going to find is that we aren’t as close as we as we thought,” Will said. “We’re going to evaluate the model and we’re going to say, ‘Oh, even though we trained it with more stuff, we’ve also now figured out how to evaluate how close to the end we are, how good the output is.’ And we’re really not as close as we thought we were.”
However, the promise of co-pilots is too great for IBM to pass up. For example, Amazon CEO Andy Jassy said in August that the company has saved 4,500 person-hours using Amazon Q to automate many of the rote Java coding and maintenance tasks. “Our developers shipped 79 percent of the auto-generated code reviews without any additional changes,” Jassy said. “This is a great example of how large-scale enterprises can gain significant efficiencies in foundational software hygiene work by leveraging Amazon Q.”
There is quite a bit of work to do on the Code Assist for RPG, however, and Will is looking for help on several fronts. For starters, he’s looking to expand his team of engineers to ramp up the AI model training, “so we can turn all of this code that people are donating to us into material that train stuff.”
IBM still needs to design a user interface for it, hook it into the IDE, figure out how to host it, and figure out how to deliver it. “So there’s still a lot to do, and that’s why I can’t commit to any of the dates that I’m talking about, because we have to ramp up the team,” Will said. “We have to do more training, we have to do more evaluation.”
With that said, Will is still open to hearing from members of the IBM i community about the Code Assist for RPG. Whether it’s sponsors, advisors, or code contributors, IBM is open to IBM i community member getting involved. IBM is also still accepting donated RPG code. The next update on the RPG co-pilot will be in February, Will said.
RELATED STORIES
Some Thoughts On Big Blue’s GenAI Strategy For IBM i
