Here’s a conversation we have more than once, in different words: “Our stack is old — Spring Boot services from years ago, a monolith we keep meaning to break up — so before we can do anything with AI, we need to modernize.” It sounds responsible. It’s usually wrong, and it’s expensive to be wrong about.
A model does not care that your service is a decade old. It needs three things from you, and none of them require a rewrite: can you get data out of the system, can you make an outbound API call, and can you deploy a small service alongside what you have. Most estates that people call “legacy” can already do all three. The age of the language is not the blocker.
What your estate has that a greenfield project doesn’t is the hard part: real data, real users, and real workflows that people depend on. That’s the substrate every useful AI feature needs. A clean new codebase with none of that is not ahead of you — it’s behind, because it has to reinvent the context your system already holds.
Legacy does bite, but not where people expect. It bites at data access — the information you need is trapped in a schema only one team understands, or spread across services that don’t share anything. It bites at observability — you can’t see what’s happening well enough to trust a new component. It bites at deployment — shipping a new service takes weeks because the path isn’t paved. Those are the real prerequisites, and every one of them is cheaper to fix than a rewrite, and useful whether or not you ever add AI.
So when the modernization argument comes up, we try to redirect it. Not “rewrite the monolith,” but “make it possible to get this one dataset out cleanly, observe this one flow, and deploy one new service next to it.” That’s a few weeks of focused work, it de-risks the actual project, and it leaves the estate better even in the worst case where the AI part doesn’t pan out.
The rewrite is the comfortable distraction — it feels like progress and defers the moment you have to build the real thing. Your ten-year-old Java estate is not the reason you can’t adopt AI. It’s most of the reason you can.