Event-Driven Agent Memory
AI agents need more than a context window. To operate reliably across long-running workflows, they need a way to preserve what happened, understand what changed, and use that history when deciding what to do next.
In this live webinar and Q&A, Tun Shwe explores how event-driven architecture can provide the foundation for persistent, auditable memory in AI agent systems, using Kafka as the event log.
The conversation will also expand into agent harness engineering, looking at the broader infrastructure surrounding an agent and how event logs can help make agent behaviour more observable, recoverable, and reliable.
What we’ll explore
Why long-running AI agents need memory beyond the context window
How event logs can act as a durable history of agent activity
Where Kafka fits into an event-driven agent architecture
Reconstructing state and context from previous events
How the event log fits within the wider agent harness
Designing for observability, recovery, debugging, and reliability
Practical architectural considerations when moving from agent demos to production systems
The session will conclude with a live Q&A with Tun, giving attendees the opportunity to discuss architecture decisions, implementation challenges, and their own agent systems.
Who is this for?
AI engineers, software engineers, architects, platform engineers, and developers building AI agents or exploring how to make agent systems more reliable and production-ready.