Agents in Production: State, Security, and the Data Layer

An official SF Tech Week session, hosted with Pebblebed.

What needs to change when AI agents become users of production databases?

Agents need to remember what happened, recognize when facts change, respect permissions, and recover when their actions go wrong. Those requirements reach beyond the model into the systems that store, govern, and serve their data.

Join Frontier Research Club for two complementary perspectives: Daniel Chalef, founder and CEO of Zep, on governed context and memory in production; and UC Berkeley researchers Shu Liu and Shubham Agarwal on redesigning data systems for agentic workloads.

Together, we’ll examine how agents acquire trustworthy context, how databases should support exploration and branched updates, and what evidence is needed to establish that these systems work.

The Frontier Research Club brings researchers, engineers, and technical founders together around papers, methods, and results. Talks are brief, with extended time for questions and critique. Papers are shared in advance so the discussion starts at depth.

Agenda

5:30pm: Doors open5:30pm – 6:30pm: Networking + light dinner6:30pm – 8:00pm: Research presentations + discussion8:00pm – 8:30pm: Networking

Presenters & topics

Talk 1 — Context as Production Infrastructure: Governed Facts, Invalidation, and the Audit Trail

Daniel Chalef · Founder & CEO, Zep

Agent memory must stay useful as facts change, permissions shift, and new information contradicts what came before. Daniel will examine how facts acquire and lose validity, how provenance and access control work in practice, and how to keep agents from relying on stale or cross-tenant context.

Daniel is the founder and CEO of Zep, the unified context layer for enterprise AI, and a co-author of Zep: A Temporal Knowledge Graph Architecture for Agent Memory. Graphiti, Zep’s open-source temporal knowledge-graph engine, is a core component of the platform. Zep’s customers include Zscaler and other Fortune 500 technology companies.

Pre-read: Zep: A Temporal Knowledge Graph Architecture for Agent Memory

Talk 2 — Supporting Our AI Overlords: Redesigning Data Systems to Be Agent-First

Shu Liu & Shubham Agarwal · PhD Researchers, UC Berkeley, Sky Computing Lab

Agents explore data through repeated queries, partial solutions, and validation—creating workloads that differ from those databases were designed to support.

Shu and Shubham will present their vision for agent-first data systems, covering agentic speculation, memory stores that retain useful knowledge, and branched updates across competing attempts at a task. The discussion will examine which database assumptions break, what existing systems can support, and where new architecture is needed.

Both researchers work on AI and systems at Berkeley and are advised by Ion Stoica; Shubham is also advised by Aditya Parameswaran. Their CIDR 2026 paper includes Databricks co-founders Ion Stoica and Matei Zaharia among its co-authors.

Pre-read: Supporting Our AI Overlords: Redesigning Data Systems to Be Agent-First

Want to present your work?

If you have a research paper you’d like to discuss at one of our next sessions, please submit it for consideration. ​Submit your paper here!

Who should attend

  • Engineers building agents that read and write production data

  • Database, platform, and security teams working on permissions, isolation, and rollback

  • Researchers studying agent memory, data systems, evaluation, and reliability

  • Founders and technical leaders building agentic products on real customer data

Capacity is limited.

We will take photos and short video clips for event recap and promotion. By attending, you consent to being photographed and recorded, and to the use of those images and clips by the organizers on social media and other event marketing channels.

🌐 Connect with Frontier Research Club

Hosted by

The Frontier Syndicate connects frontier technology researchers, builders, and investors through curated research forums, private dinners, and early-stage capital. We back exceptional companies emerging from the labs, communities, and technical networks we convene.

Pebblebed is a technical early-stage venture firm founded by Pam Vagata, co-founder of OpenAI; Keith Adams, founding engineer of Facebook AI Research and former chief architect at Slack; and Tammie Siew, former Sequoia Southeast Asia investor and venture-backed founder.

Constructive builds secure-by-default Postgres infrastructure for agents. Versioned, composable modules bring schemas, policies, functions, and roles together, with row-level security and multi-tenancy enforced at the database layer.