How to Decide What Your AI Agent Should Remember - Wojciech Pluta, Oracle

By ai.engineer

Categories: AI, Tools

Summary

Memory in AI agents isn't just conversation history—it's a write path that requires deliberate "memory promotion" pipelines to classify, compare, and validate what's worth storing. This data layer is where competitive advantage lives, not in commoditized models.

Key Takeaways

  1. Distinguish retrieval (read path) from memory (write path): RAG systems retrieve but don't decide what survives interactions. Build promotion pipelines that classify candidates, compare against existing records, and admit/reject based on evidence requirements.
  2. Implement a 3-stage memory promotion pipeline: (1) classify meaning of candidates, (2) compare with existing records and check evidence, (3) harness decides admit/reject/escalate to human review. This prevents both data bloat and memory loss.
  3. Data layer ownership is your competitive moat, not models. Models and infrastructure are commoditized; your proprietary data and how agents learn from it drives sustainable business value in financial services and beyond.
  4. Move from retrieval-augmented architectures to intent-based memory systems. Agent harness owns runtime decisions around what to keep; this enables just-in-time, on-demand software that creates continuity across interactions.
  5. In production agents (mortgage briefs, document processing), explicit memory schemas prevent missing critical documents. Pre-recorded demos showed agents flagging missing mortgage docs—memory classification catches business-critical gaps retrieval alone misses.

Related topics

Transcript Excerpt

Uh my name is Wojtek. Uh Wojtek Pluta, and I work at Oracle. And over the last 10 years, I've been working um at Oracle. I worked at the various projects um including working with some banks in the UK on projects like customer experience mortgage use cases or things like early contactless payments when um when it was still uh sort of emerging and open banking initiatives. Now, I am uh heading up the technical marketing team uh at Oracle AI database. And before, I spent a bit of time also um working on cloud native and AI on our cloud infrastructure. So, what I'm going to talk about today is the um the moving from rack to memory systems and the notion how rack gives us a useful read path, uh but memory system allows us to uh create unique business value based on your data. So, over the year…

More from ai.engineer