AI agents don’t slot neatly into the way enterprise data stacks were designed. A typical agent needs transactional state (e.g., what just happened), long-term memory, similarity search, relationship-aware data, and the ability to react to data changes in real time. In most organisations, that means stitching together a relational database, a vector store, a graph engine, and layers of middleware to keep everything in sync.
“AI agents don’t slot neatly into the way enterprise data stacks were designed.”
That architecture can work, but…
