AI agents are starting to behave less like stateless chatbots and more like production systems: they call tools, mutate state, personalize behavior, and return to the same user or workflow over time. That creates a new reliability problem. When an agent remembers the wrong thing, the failure is subtle: future behavior changes, the prompt still looks reasonable, and the team may not know which prior read or write caused the issue. This beginner-friendly talk looks at agent memory from an SRE lens. We will walk through common failure modes such as stale memory, accidental overwrites, unbounded personalization, hidden prompt state, missing rollback, and poor auditability. Then we will discuss practical controls: explicit memory tools, scoped user/shared memory, revision history, access logs, safe write boundaries, and simple operational checks. I will use MemexAI, an open-source Postgres-backed memory layer for agents, as the demo system, but the focus is on reliability patterns teams can apply to any agent stack.
Attendees will learn:
Sooraj Sanker is a Bengaluru-based fullstack and AI engineer, and the founder of MemexAI, an open-source Postgres-backed memory layer for long-horizon AI agents. He spent roughly six years at Hasura and PromptQL, where he worked on conversational analytics, AI agent interfaces, tool-based workflows, artifact-aware chat, and state management for distributed agent execution. His work sits at the intersection of agent infrastructure, context engineering, data UX, and production reliability.
