Can AI Agents Handle Complex JD Edwards Customizations?

Does having a highly customized JD Edwards environment make adopting AI more complicated? Not necessarily. In this episode of Not Your Grandpa’s JD Edwards, Nate Bushfield and Manuel Neyra explore how AI Agents can work alongside existing customizations, integrate with your current systems, and automate business processes without requiring a complete overhaul. Discover why your unique JD Edwards environment doesn't have to stand in the way of AI innovation.

Does AI sound great in theory, but unrealistic for your highly customized JD Edwards environment?

And one of the bigger issues isn't custom code, but a business process that only makes sense inside your company.

Today, we're breaking down what customizations actually matter when it comes to AI agents.

And by the end, you'll know how to tell whether or not one of your unique processes is a good AI candidate.

Welcome back to Not Your Grandpa's JD Edwards.

I'm your host, Nate Bushfield.

One objection we hear all the time is AI sounds great, but you don't understand how customized our environment is.

Sometimes that means custom applications and integrations.

Other times that means the business process itself is unique.

So today, we're asking a simple question.

Does being highly customized actually make you a bad candidate for AI agents?

Joining me today is Manuel Neyra, VP of AI and Products here at ERP Suites, to help us break it down.

Manuel, welcome back to the show.

I know you're a recurring guest.

You probably have five or six episodes at this point, if not more.

But for the people out there that maybe haven't seen an episode with you, could you maybe give us a little bit of background?

Yes, first of all, Nate, always good to be here with you.

Always a good time and we always have interesting conversations so looking forward to today's session.

For folks that don't know me, you know, I, I come from a long history of product development all the way back into the mid-90s with the JD Edwards organization through the PeopleSoft acquisition and Oracle, of course, adding up to almost 27 years being there now.

I'm, I'm happy to be here at ERP Suites leading the, the AI team and the investments we're making, both from an agentic perspective as well as a platform perspective to allow customers to safely and functionally have agents that drive efficiencies for their business.

So really excited for today's topic.

Yeah, I know it's, it's something that we've danced around.

You know, we talked about customizations, we talked about how specific some customers can be, but we really haven't dove deep into this topic.

I mean, how we often hear AI sounds great, but our JD environment is highly customized.

Is that actually a barrier or is that just something that people like to say because they think they're different?

It's, it's definitely the latter, Nate.

It's, it's definitely because they think they're different and many companies are different, right?

What you know, you look at the different verticals of companies manufacturing chemicals, even med device companies, oil and gas are looking for a differentiator, right?

And those differentiations get implemented in JD Edwards and sometimes it's supported by base software that JD Edwards delivers in some cases and not, it's not, right?

It adds additional capabilities that doesn't prevent customers from using AI.

It adds additional considerations for sure.

And all the tenets that we've discussed maybe in prior podcasts, Nate, where identifying the right process and making sure that it's a defined process, it just puts a little bit higher premium on customizations to make sure that those special sauces or special business processes that make that company different are incorporated in AI.

Business Processes vs. Technology: What Makes Customizations Challenging?

Yeah. So that last point there of it being business process, would you say that this is more of a business process problem or could it also be like a technology problem?

That's a great question, right?

I think I would say 90%, 10%, right.

Most of it, it's, it's a business process challenge, right?

And making sure that it's documented, make sure that it's accurate in terms of what you know, that you know, whether it's that special sauce or the mainstream type of scenarios, exceptions as well are documented is going to be tantamount, right?

The 10%, you know, there, you know, as I reflected on, on this topic, the 10% could be made up of scenarios where companies become very, you know, ingenious in terms of how they implement particular customizations, right?

It's one thing to use orchestrations, it's one thing to use logic extensions or any of the constructs that JD Edwards has, right?

Even the, you know, the, the, the, the long legacy type of ways of adding a new application and so forth supported.

Then there's really creative other ways that then become very intricate and maybe have third-party type of technologies embedded in JD Edwards.

And it's a little bit of the beauty of JD Edwards, right?

You can, you can use different technologies.

Those don't necessarily make it prohibitive for you to implement AI.

It just becomes a little bit more complex in terms of how you're going to interface to a third-party technology with third-party data and how do you bring that in.

As we all know, there's different constructs for doing that, whether it's API calls or MCP, right, that everyone's talking about.

So there's, there's different levels, but it, it adds a little bit more layer of complexity into the process, but doesn't make it prohibitive.

Yeah. But even then, does every AI agent need to understand every customization in the environment or are there some exceptions there?

No, great point, great point.

It all goes back to what is the customer looking at trying to realize as a result or the benefits that they want and being able to identify that includes assessing the, the flow of the business process, right? And what is relevant and what is not perhaps for what they're looking to accomplish.

So that, that is extremely key to your point.

But, but it, but again, I, I go back, right.

Sometimes seeing an exception doesn't necessarily mean that, you know, you can't adopt AI.

Yeah, yeah.

So obviously we can talk about third parties and that type of complexity there.

And yes, I get how there are customizations.

Maybe there's some added security that you might need to add on here and there because of the different levels between each application.

But what if the business process, like the actual business process that they are trying to perform with this AI agent is completely unique?

Is there another level of complexity that an AI agent will have to go through or is it a teaching moment?

Can you maybe unpack that a little bit?

That's a great question.

And there are some interesting things in my career, right, that I've seen of what customers have done in terms of needing to customize JD Edwards to make it work specifically for what they're trying to accomplish for the business process.

And sometimes, right, it was, it was above, you know, 60–70% kind of customization, some significant changes.

And from, from that perspective really it's understanding the functionality, right.

So our agents, for example, what we're building at ERP Suites, you know, we are building some agents that are off the shelf that are kind of catered to base JD Edwards functionality.

But our, but our architecture allows for and expects some specific type of processes that need to be implemented for a customer.

Now, if a customer is highly customized, we can still tackle that and we can still address the business process that they have, even if it's 60–70% customized, it, it takes a little bit more work.

But we're also anticipating that, Nate, in our architecture so that if the base model needs to reflect a customized JD Edwards application or process, we can do it efficiently rather than having to tear everything down and bring everything from the ground up.

And that's oftentimes when the question either comes up or where we're talking right now is, well, you can't use AI because it's impossible.

Well, you know, slowly we'll, we'll talk about that some more here, but the, the other point is, well, is it going to be costly?

Is it going to be cost prohibitive?

Am I going to invest more in the agent?

Am I getting the return?

And that's why, you know, I, I, I feel confident that what we're building allows for that balance so that you can still have your, your, your customized JD Edwards and still get the benefits of AI that everyone's looking for.

See, that's an interesting thing you bring up there of realizing the ROI because of how customized maybe your environment is.

Thing is there, like it's kind of like you already have to teach an AI agent your process anyways.

So if it is a little bit more customized, you're already going to have to teach them the process.

So what's a few more steps?

You know, I mean, obviously you're going to know that answer more than I do, but yeah, so like it.

Yeah, sorry, go ahead.

No, no, no, I think you hit the nail on the, on the head, right.

It's, it's about training the, the, the model, setting up the model for the agents so that it understands the process that it's going to be monitoring, assessing or leveraging to complete a set of tasks.

As long as you do that, you're, you're going to be in big, big shape.

There's nothing prohibitive from it, right?

There are some, you know, I'm not going to overblow it, right?

There are some things that will cost a little bit more time or effort to do, but prohibitive.

I think it's, you know, maybe what's prohibitive, right, is if someone doesn't have, like we've spoken in the past, clearly documented processes, clearly, you know, paths that people, humans today run, if there are exceptions and they're either tribal knowledge or even worse, right, that there's no tribal knowledge, right?

And it's been lost, obviously that's going to cause problems, not just with AI, but with the system itself.

So it's kind of a precursor, right?

You need to have some of the basics in your ERP, JD Edwards or any ERP before you start, you know, layering in AI.

So yeah, it's, it's the agents only, only need to understand part of your environment tied to its job.

That really narrows down the actual problem because, yeah, where you can talk about third-party applications that might be plugged in, but if your AI agent is never going to be a part of that process anyways, it doesn't really need to understand that.

It can just work on its side of it.

It's kind of like a, like a regular person in that way of, yeah, you can have all these customizations over on the finance side, but if someone on the marketing side is working in JD Edwards for whatever reason, they probably won't need to know exactly how to utilize those third-party features because they're never going to work with it.

So that makes sense.

So what types of JDE customizations actually matter when implementing an AI agent?

I would say the ones that, that matter, right, are probably across the board in terms of the capabilities, right?

If they're contributing an additional step in a process that requires an approval or is capturing additional data or is aggregating data based on a selection or a decision that a human is making, those are going to be relevant, right?

Because now the agent needs to understand how to process that, how to augment it, how to analyze it differently with, if there's a new data element or a, the collection of data now that is being fed to enrich, let's say, for example, visibility on financial planning that, you know, company didn't have before.

It's going to be essential to have that.

So I, I think it really, you know, goes back to what is, what it, what is critical to the process and, and, and incorporating it in the model, training it and making sure it's, it's good.

If the back, it goes back to what you were talking about earlier, if the particular, you know, set of customization is not in scope and let's say they made a table change, it, you know, it's, it's, it's, you know, there may be some mapping that needs to happen, but that's a drop in the bucket.

And, and then within our model and how we're building agents to better update that, but tangible kind of relevant impact to, oh, now we need to make sure we consider this in terms of a customization, whether it's localization.

If we start talking about the international world where there's localizations that need to be accounted for, that's going to be very relevant, right?

So those types of customization maybe that JD Edwards does not deliver, but that, you know, either the customer has developed or partners developed localization for, you know, whatever, right, taxes in a certain part of Latin America.

It's, it's, it's going to need that and it will be, it will take some time.

But does that, is that prohibitive?

Again, I keep on beating that drum.

It's, it's not.

And you kind of touched on it right there.

But just so we're clear about this, can an AI agent work with custom JD applications, tables, business functions or even orchestrations?

Or do we have that issue anymore?

That's a great question.

No, we don't have that issue, right?

That's the beauty of what JD Edwards built, what we built back in the day, right?

It was at JD Edwards, is we built, we had a tool that we used to create applications, tables, business views, reports.

And now with the, with the, with the, you know, extension capabilities, right, of, of user-defined objects and all the feature sets that, that JD Edwards team has done a marvelous job of adding.

It gives you that expansion.

So of being able to leverage those capabilities.

So as a customer, if you're adding these components, right, as customizations to your base objects, or in some cases those are personalizations, it's still something that JD Edwards recognizes, right?

So it's, it's an object.

So when AI is laid on top of it, right, it knows, it will know about user-defined objects, it will know about tables, it will know about business views, etcetera, etcetera.

So in some sense it doesn't care, AI doesn't care whether it was something that was delivered by the JD Edwards team or if it was delivered by a partner or by the customer, right?

It's, it's, it's kind of like the same toss and in the same way in some ways it's indistinguishable for the agent.

Evaluating Customized Environments for AI Readiness