MTS Why LLMs Need a Collective Memory Layer | Pablo Stern-Plaza
9 snips
Oct 5, 2026 
Pablo Stern-Plaza, MongoDB’s Chief Product Officer for AI and Engineering, explores the infrastructure behind production-ready AI agents. He discusses Atlas Agent Engine, persistent memory, enterprise security, local and hybrid models, and how multi-agent systems coordinate through shared data. The conversation also examines data gravity, intent-driven skills, and how agents may reshape work.
AI Snips
Chapters
Transcript
Episode notes
MongoDB Packages Agents Around Memory And Governance
- Atlas Agent Engine combines production deployment, a governed memory layer, and flexible execution environments for AI agents.
- MongoDB positions it as an end-to-end stack rather than another isolated agent component.
Customers Outgrew Markdown Agent Memory
- Customers told Pablo Stern-Plaza that MongoDB was a natural home for agent memory because it combines JSON documents with text, vector, embedding, and reranking capabilities.
- Many teams had prototyped with Markdown files or wikis but needed production-grade controls once memory became business system-of-record data.
Production Agents Need Security And Deployment Choice
- Enterprise agent adoption depends on centralized security, role-based access, workload isolation, and deployment flexibility.
- MongoDB’s ability to run across clouds, on-premises, or air-gapped environments addresses regulatory, cost, locality, and sovereignty concerns.