IBM i DevOps TechTalk #28
AI – PART 1

Ray Bernardi – Welcome to IBM i DevOps TechTalk, where we discuss key topics and questions with ARCAD experts. I’m Ray Bernardi and I’ll be your host today. Now here at ARCAD, we’ve been studying and learning all about AI, how to use it, how to use it safely, how to use it securely. So joining me today is Alan Ashley. He’s basically our resident AI expert. And I’m going to be asking him some questions about AI. So Alan why is AI for the IBM i such a hot topic right now?

Alan Ashley – So when it comes to AI, it has become this exponential growth not only for the i, but in all forms of technology. As you can see, day to day all the news reports. But on the IBM i side, it’s really being driven by the advancement of the models that understand the IBM i. And some of this is coming from the IBM Bob project, where they’re building AI into the IDE. That’s going to connect directly into the IBM i.
So that’s part of it. The other part of it is really being driven from the top that all the executives they want AI in their development. They don’t want to come to their peers and go, oh yeah, we’re not using AI because they’re afraid they’re going to get left behind. So the developments are coming in.
And what’s really nice about this is everybody is on the same page at this moment when it comes to AI, in that it is so new for all the platforms that we’re all learning it together. And what’s nice is we’re finding the value in the use cases that are being developed around the IBM i.

R.B. – Okay. So Alan, you mentioned use cases. Can you elaborate on that just a little bit. What kind of use cases are you talking about?

A.A. – So one of the great things about AI is this uncanny ability to recall, meaning the models have been trained on all this data and all this knowledge, and it’s instantly available. So when you’re looking at your source on an IBM i, maybe it’s an RPG fixed, maybe it’s RPG III. You’ve lost some of that human knowledge on how that functions.
AI has all that in this repository of information. So it’s literally there. So you can ask it questions about fixed format. You can ask it questions about modernization.

R.B. – So there’s no such thing as arcane knowledge to AI.

A.A. – Exactly. So there’s no such thing but you can also have too much information. Meaning information overload. And so that can even confuse it. And this is where, we have tools like DISCOVER and our MCP that can fine tune that information as you will, but we’ll get to that in a little bit.

R.B.- When you say DISCOVER and MCP, you’re referring to ARCAD tools.

A.A. – Yes, ARCAD tools of DISCOVER, which has an AI component into it. And it allows you to really fine tune how your application is viewed to the AI and MCP. We’re all familiar with how that works. And basically the handshake between the knowledge and the factual information on your systems in the AI to keep it deterministic. And so those are a couple of ways you can approach those things.

R.B. – All right. So you mentioned basically it almost sounds like a double edged coin. You know it has arcane knowledge. It can tell you everything from the past. And however it can be all a lot of information, maybe too much like taking a sip from a firehose, if you will. So how do you prepare for this?

A.A. – So one of the things to do is and this is the way I approached it, and things change so rapidly around this topic that what the approach is today may not be valid tomorrow. But today what I did is I went out there and started asking it questions and finding the use cases that I needed solved, the problems that I needed solved.
And one of the things is, is when you’re preparing to make the move to AI on your own IBM i, find the problems that you need it to solve. Because if you just go into it saying modernize this, you’re going to get very different results as opposed to if you go through a context of using an MCP, maybe supplying some RAG information through documentation and a vector DB.
Now there’s no add to be there on day one. It can be day two or day three when it comes to things, but find your use cases, otherwise you won’t find the value in it. Other than saying that you’re using AI, that’s just my point of it. And this is what I tell people when I talk to them and comment this like, okay, we’re looking at going with the AI, we’re going to go with cloud. We’re going to go with OpenAI. I use Codex, I said, okay, that’s good. What problem do you want to solve?

R.B. – And they go blank.

A.A. – And they go blank because they haven’t made it that far in their discussions. And to me, start with this documentation. Start with the documentation of your application using AI. And one of the things you’ve probably noticed is that sometimes that documentation might be a little lacking, or it might be a little fluffed. So to speak, because the AI doesn’t understand the context. That’s where an MCP like the ARCAD MCP comes in, and it can provide the context and the true nature of relationships between components within your application. So you start to build that confidence in your AI and how the AM models working with your information. That’s really key. One you got to find your use cases. Two you got to build the confidence within yourself of what the AI is providing.
And then you had to really double check what the AI is providing. Because at the bottom of every AI model, it says: AI can make mistakes, double check the results.

R.B. – If you say AI can make mistakes that brings up a whole new question. There should be some concerns here then. What kind of concerns are there? What should people be worried about when using AI?

A.A. – So there’s a couple of different things in it. So you have the training of the models and there’s always the option of allowing your data to be used for training, meaning it’s going to take what you provide and help train the models. That’s a concern. Maybe your source code is being exposed. That’s one concern. So a lot of people are looking at going to private models, meaning it’s on their prem and it’s not being shared with the rest of the internet.
Another way to look at this is how your developer access comes into it. Meaning are you going to a developer for the vibe code? Your application and push it all the way to production? No, you don’t want to do that. You still need validation along the way. We’re not to that point yet, where you can develop a full blown application, push it all the way to production, and have it work flawlessly.
That’s not there yet. And again, I preface this with yet because the advancements that I’ve seen in the last six months have been just staggering to someone a Gen Xer. It’s really been staggering how quickly it has advanced. And so there’s always those security concerns. And this is where an MCP like the ARCAD MCP comes into play in that we’re providing a true secure information.
We’re limiting how the AI can see the information. We’re only providing what you asked for in the information so that you know that when you go out, you have a higher confidence level going into it and coming out of it with what is being pulled.

R.B.- So to wrap this up, Alan, what would you say that? I mean, you’ve mentioned MCP a couple of times and so on. So is that really what ARCAD is bringing to the table?

A.A. – ARCAD is coming at it from a couple of different ways. So the big one is the MCP which is free if you’re an ARCAD customer. So you can download it right now it’s in tech preview as of the may recording of this and the GA release is coming shortly. So that’s one area that we’re bringing AI into this, and it’s providing that deterministic result through tools and looking at APIs within ARCAD.
But we also are using it within our DISCOVER product that can go through and do a little bit of code explanation. It’s pulling in some of the version history into the documentation aspect of it. And another area that we’re going into now is anonymization aspects of it. So within our anonymizer tool, we’re actually using AI to help develop the groovy scripts to help you anonymize your data. That’s a good use case right there. And that one was really exciting for me to play with and see how that came together.

R.B. – That is a good use case. As a matter of fact, we just completed a video on that, and that should be available out there on our site somewhere this point.

A.A. – Yes. And that was really fun to play with because you could ask it a question, a human language question, and you got the formula in the script of the groovy script that you’re going to apply to your anonymization. And what’s really nice about that is when you’re asking it as a human, it’s now harder to reverse engineer the script because everybody’s going to ask the questions differently.
Therefore, the scripture are all going to function differently, meaning you’re not going to have the patterns available. They can be reverse engineered. So that’s really going to be something really exciting along those lines.

R.B. – Alan, we’re starting to hit a little bit of time limit here. So I think we’re going to have to do another techtalk on this. We’re going to have to bring up a part two on this AI thing and talk about this again in the future. What do you think?

A.A. – Oh, I think definitely. And some of the advancements that I’ve seen in our roadmap and in our industry will definitely be a part two and probably a part three.

R.B. – All right. That sounds good. Let’s wrap it for now then. Alan, and thank you for joining us today.
So that was part one of our AI discussion. Now we thought we would do a techtalk on AI and be done with it. But obviously there’s a lot more to discuss. So stay tuned and look forward to the next tech talk.