Most frameworks—like OpenAI Agents SDK, Google Agent Development Kit, and LangChain—run agents in memory. When the process dies, so does the agent. Fine for demos; fragile in production.
Agentspan is a durable, distributed runtime for real-world AI systems. Instead of disappearing on failure, agents persist, scale across machines, and can pause for human input—even for days.
This talk shows how durable execution turns brittle workflows into reliable, resumable systems. You’ll see why in-memory agents break outside controlled environments, and how to run crash-resistant executions, scale across workers without rewrites, and implement multi-agent patterns that actually hold up.
We’ll also cover long-running, human-in-the-loop workflows and practical ways to adopt Agentspan with minimal changes to your existing stack—giving you a clear mental model for why durable execution is the missing layer for production-grade AI agents.
Speaker
Viren Baraiya
co-founder and CTO @Orkes
Viren Baraiya is the co-founder and CTO at Orkes, a leading workflow orchestration platform and maintainer of Conductor, the open-source project used by Netflix to achieve reliability at massive scale. With over 20 years of experience, Viren has led engineering teams at Google, Netflix and Goldman Sachs focusing on building developer platforms. He was also lead software engineer at McGraw-Hill Companies where he led the design and implementation teams for the US and India, building some of the industry’s first online edTech products.
Session Sponsored By
The Modern Workflow Orchestration Platform - Seamlessly build and run complex workflows, applications, and integrations without the hassle of building infrastructure - all on Orkes.
Video is not available
