Why your agentic AI keeps failing in legacy environments - and what the architecture actually has to look like - London Business News | Londonlovesbusiness.com

"The first time an agentic AI deployment failed at scale, it wasn't because the model was wrong. It was because the model couldn't see anything that mattered."

"Twenty years of business logic had lived inside the legacy stored procedures. The migration had moved the schema. It had not moved the semantics."

"Most teams reach for retrieval-augmented generation when they hit this wall, because RAG is the familiar tool. It works beautifully for documents - a haystack with a needle in it."

"Legacy code is not a haystack. It is a graph. A monolithic codebase is a dependency network where touching one function ripples through dozens of others in non-obvious ways."

Agentic AI deployments often fail due to architecture problems rather than issues with models or prompts. A case study illustrates that after migrating a legacy database to a cloud-native system, the new agent could not reason across business contexts. The migration preserved the schema but not the underlying semantics of business logic. Many teams mistakenly use retrieval-augmented generation (RAG) to address these issues, but RAG is ineffective for legacy code, which functions as a complex dependency network rather than a simple document structure.

Read at London Business News | Londonlovesbusiness.com