Wojciech Pluta, who leads technical marketing for Oracle AI Database, detailed a memory promotion pipeline to decide what an agent keeps on AI Engineer.
Retrieval is the read path and memory is the write path, Pluta said, with promotion deciding which candidates survive the session.
Pluta said "models are shared but memory is yours" to argue the data layer, not the model, creates durable advantage.
His demo built a mortgage brief for Kyle, stored a semantic preference, then reused it for Vincent while blocking a blanket approve-every-application request.
Pluta said the harness classifies candidates, admits or rejects them or routes for human review, and persists admitted records in Oracle Database.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 12,007 subscribers.
Oracle published notebooks on GitHub and a revamped video series to show the agent memory package introduced earlier this year, which still leaves token budgets and provenance policies to operators.
