RPG Code Assistant FAQs: Steve Will Gives Answers
As part of IBM's Guided Tours series, the CTO of IBM i responded to some of the most common questions about the RPG Code Assistant project
Last month, IBM i CTO Steve Will provided a highly anticipated update on the progress of the RPG Code Assistant (RPGCA) project that was announced last summer. Via livestream, Will described the three main functions that his team aspires to include in the code assistant—explain code, generate code and write test programs—and announced plans to release the RPGCA beta in Q2 of 2025.
Additionally, he answered several frequently asked questions (FAQs) that had been posed by an eager IBM i community. His answers to those questions follow.
First, The Easy Ones
The following answers to the RPGCA FAQs come with Will’s caveat, “as far as we know today (Dec. 11)”:
Q: What large language model is going to be used in their RPG code A system?
A: “We’re certainly going to start with a Granite model and train from there.”
Q: Will RPGCA be able to convert fixed-format RPG to free-format RPG?
A: “While it’d be nice if our assistant could do that, there are other tools today which, without using AI, do a very good job of this. So this is not something we’re trying to replace. But we are teaching the model both flavors of the language.”
Q: Will the code assistant convert RPG to COBOL?
A: “No, that is not a goal of the RPG Code Assistant. We want to be able to help people hire people to work with the RPG that they have, not convert it to some other language.”
Q: Will there be a COBOL Code Assistant?
A: “While we have heard this request, we’d like to focus on it once it’s clear that the RPG code assistant is meeting the needs of the RPG user.”
Q: “If I am an advisor, will I get an early version of this?”
A: “No earlier than beta for sure.”
Q: Will there be support for natural languages other than English?
A: “It’s not clear if we can accomplish it in the first release or not…depends on how we design this thing and what resources are available to us.”
Q: How good is the output going to be?
A: “The output, I’m pretty sure, is going to be pretty good in explaining, but that’s not what they’re really asking. Generally, they’re asking about a code assistant replacing developers. Is it going to be good enough to replace a developer? No. Code assistants are about assisting developers to deal with code that is unfamiliar to them, perhaps in a language that’s unfamiliar to them so that they can be accurate and productive.”
Q: Will you support an Eclipse-based client compatible with RDI?
A: “We know there’s a large segment of our user base that uses RDI. We know this is a key requirement. However, we’ve got to develop with something first. VS Code is not only more nimble to work with, it’s 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 as something that we would have to do later and prioritize it in with other things.”
Q: What are you doing with the code that people contribute?
A: “It is stored in a private repository. Next, the team checks the code’s legal status to ensure they can use it. Upon confirmation of the legal status, the two parties sign an agreement and the team makes sure the code itself doesn’t have extraneous stuff that we have to worry about. Then, some of the code is turned into training pairs consisting of inputs and their corresponding outputs.”
“We’ve asked our contributors if they can help us create those training pairs. And some of them are doing that, and I really, really appreciate that. Eventually, some of the contributed code [depending on privacy stipulations] is going to get put out into a public Git repository so that we can show people some of what we trained with.”
Q: How do I get involved?
A: “If you do want to get involved, there are several levels. One level is as an advisor who will be contacted with questions about how they want to consume the product and what functions they would like to see. One level deeper in involvement is the sponsor user, who will be contacted several times during development. Another way to be involved is to contribute code for the model training. This is the type of participant who will receive the Alpha release of RPGCA.”
Q: Whom do I contact?
A: Contact AIforIBMi@ibm.com and wait for a response. “And then, depending on what you want to do to contribute code, you’ll have to agree to the license, submit the code in a way that we’ve got described in the GitHub repository and so on.”
FAQs Without a Clear Answer
Some of the FAQs that Will has been fielding don’t yet have definitive answers, such as the following:
Q: How much is it going to cost?
A: “We don’t know. We know that our client base, especially at the low end, can’t afford a super expensive thing, and we want this to be quite available, and we are taking that into account. But we also know that there are larger clients who have a lot of people who have to use this, and they have more extensive requirements.”
Q: How will you charge for it?
A: “We’re not sure yet. It could be a subscription, it could be usage based…We are looking at advice from our advisors on what our client base wants.”
Q: Where will the model be running when we use it?
A: “At the beginning, the model will be running in an IBM cloud and offered as a service.”
Q: Will the model be open source?
A: “Unknown at this time.”
Q: Will you API-enable this to allow other business partners to extend it?
A: “We’re unsure, but please be aware, I know that the IBM i ecosystem that we have around us is critical to the success of IBM i, and there are things that our partners do that are specific to their solutions, installations, tool sets, et cetera. I want to be able to make it possible for those people to do their own specific things and take advantage of this model and the offering that goes around it. So we are investigating how we can make both of those things happen in the right way.”