Is Agentic Modernization Finally Possible? 

Nobody has done agentic application modernization a hundred times yet, because the AI models capable of it have only recently become good enough to use. LiminalArc’s bet starts with a thesis: business capability modeling and domain-driven design are the same discipline in two languages. One extracts and encapsulates the business, and the other extracts and encapsulates the technology aligned to it. Combined with AI extraction tools and context engineering, that thesis supports an extract, enhance, exploit cycle for legacy systems.

The approach rests on 16 years of LiminalArc transformation work, including six years of extraction and modernization done by hand. That history is where the judgment about keeping humans in the loop comes from. Agentic practices are already speeding up that work on client engagements, alongside agentic playbooks and agentic governance. It isn’t proven end to end yet, but it’s close enough to talk about openly.

Video Transcript

Mike: We’ve been doing extraction and modernization work for six years by hand. We know the first principles of doing all this stuff, and we know where humans and judgment need to be in place.

And now with the advent of AI, over the last year and a half, we’ve been building the pieces into the client engagements that we’ve done.

Stacy: Well, I do love how you guys anchor things in patterns. And so if we go back to talking about AI, knowing that it’s the hot topic across every boardroom, what are you betting on as it relates to AI?

Mike: We started developing this thesis about two and a half years ago, started investing in it about a year and a half ago. About nine months ago, six months ago, the models started getting good enough to do some of this stuff.

There’s a lot of folks out there that are doing application modernization with AI. We didn’t invent that little pocket. But what I think is unique about us is an insight I had, probably one of the first things I explored with ChatGPT when I got my ChatGPT-4 account back in the day.

The thesis was: is domain-driven design and business capability modeling the same fundamental discipline in two different languages? Business capability modeling being a framework for figuring out how to extract and encapsulate the business, where domain-driven design is really about how to extract and encapsulate the technology in alignment with the business.

And what was funny is that as I’m querying AI, it’s fighting my thesis the entire time, but in fighting my thesis, I kind of went, “I’m right.” And then I spent the next six months enlisting the rest of the organization to see it the way that I see it.

What am I betting on? I’m betting on that thesis. I’m betting on the idea that that thesis, combined with AI extraction tools and the idea of context engines, an extract, enhance, exploit cycle, where you extract the business capability, you align it to the domain, you enhance it so that you can do really clean agentic development, you can run really clean AI use cases within that encapsulated component, and you can start to expose that data in a way that’s available to other parts of the organization.

So the exploit side of it is that once I have the encapsulated components, I can start to have agentic workflows going across them. Then the next step of that, and this is the stuff that’s super hard to talk about with everybody, because that’s not where everybody’s head space is.

That starts to imply agentic extraction and modernization. It starts to imply agentic change management. It starts to imply agentic SDLC. It starts to imply agentic systems of delivery.

And then what’s going to be required for all of that is: where do humans fit into that, the humans in the loop, where do they fit and where can they be taken out? And to even begin to have that conversation, I think this conversation that is happening in some places in the market, but I don’t think is super well understood, is context engineering. I think that’s a really, really big deal.

We’re piecing all that stuff together, and that’s what I’m betting on.

Stacy: I know you said that this is your thesis, but have you actually seen it start to work in any engagements that you guys are doing today?

Mike: It’s funny. I pay a lot of attention to X and Substack and different things, and I’m trying to figure out where people are talking about what they’re doing. There’s a couple articles I’ve read that have really kind of lit up my brain.

This one article was talking about the idea of going to CIOs or CEOs and saying, “Where in your organization do you have a hundred people doing something where two people could do it?” And then this one person talked about this idea of, “We’ve done this a hundred times.”

I just went, “I don’t think anybody’s done this a hundred times.” I’m sure they’ve done whatever they’ve done a hundred times, but in our context, nobody’s done this a hundred times. Technologies didn’t exist six months ago to do some of the stuff that we’re doing.

The case I have to make is that LiminalArc has 16 years of doing this by hand. We’ve been doing extraction and modernization work for six years by hand. We know the first principles of doing all this stuff, and we know where humans and judgment need to be in place.

And now with the advent of AI, over the last year and a half, we’ve been building the pieces into the client engagements that we’ve done. We have clients that we’ve done ingestion and extraction work on their code bases using something we call Code Navigator. As we’ve extracted and modernized different applications by hand, we’ve started to bring agentic practices into that to speed up that work.

We’re doing a lot of stuff with agentic playbooks internally. We’ve started working with some of our customers on agentic governance. And all these little pieces that are emerging in the marketplace are starting to snap to grid.

What’s kind of interesting is that because we’ve got this integrated methodology, we’ll use AI and these techniques for the places where they’re mature, and we still have the background and expertise to do it by hand in other places. We know the pitfalls in all the different areas, because we have so many reps doing it over the last 15, 16 years.

So where have we seen it actually work? We’ve seen it actually work by hand all over the place. We have some really solid use cases going right now where we’ve saved and made clients lots of money implementing these things, and they’re just becoming more and more identified over time.

Have I totally seen it work end to end? Nope. But I think we’re close enough to really start talking about it and share what we’re learning. There’s going to be some people that are going to go along for the ride with us.