Date: Barcelona · 22–23 September · Booth E4
Most AI agents hit the same wall in production: they can’t remember. Conversation state sits in one system, embeddings in another, business records in a third. Every retrieval becomes a sync problem, and every sync problem eventually surfaces as an agent that contradicts itself.
The teams that got past this didn’t reach for a bigger context window. They replaced the layer underneath with a single database that serves vector reads, ACID writes, and analytical queries in the same transaction. No sync pipeline to keep honest, no drift between what the agent retrieved and what the system of record says.
Stop by Booth E4 and we’ll walk through the architecture with you: how agent memory maps to tables, what multi-tenant isolation looks like at scale, and where TiDB is the wrong choice for your workload.
On Stage: Breaking the Memory Wall
23 September, 13:40
Bernard Kavanagh, Principal Solutions Architect EMEA, TiDB
Bernard breaks down what the memory wall actually is, why it caps agentic systems in production, and the architectural decisions teams make to get past it.
Bernard Kavanagh, Principal Solutions Engineer at TiDB
Bernard is a specialist in high-performance data systems, helping organizations rapidly pivot from disjointed data stacks to unified infrastructure required for the modern agentic era. At TiDB, he architects Agent-First memory stores that consolidate Vector Search, Hybrid SQL, and Real-Time Analytics into a single platform , replacing fragmented multi-service stacks where AI agents lose context and accumulate cost. His work productionising AI workflows includes a real-time multi-agent platform for 200,000+ IoT devices that cut infrastructure costs by 65% and LLM token consumption by 10x with agents that improve diagnostic accuracy from every investigation.