As AI agents become first-class consumers of open-source data platforms, familiar reliability patterns start to break down. This talk looks at failure modes around changing data, dynamic execution, side effects, lineage, and replay, and explores the infrastructure patterns needed to make agentic systems reproducible, debuggable, and recoverable.
Siddharth Arun is a Senior Member of Technical Staff at Salesforce, where he works on large-scale distributed systems, data platforms, and AI infrastructure. His experience spans enterprise data migration, cloud platforms, observability, security, and production AI systems across Salesforce, Broadcom/VMware, Exabeam, and InMarket. He is especially interested in how open-source data infrastructure is evolving to support agentic AI, with a focus on reliability, lineage, observability, and recoverability in production systems.