From Stateless Nightmares to Durable Agents - Building Production-Ready AI Systems with PydanticAI and Temporal

Learn to build production-ready AI agents that survive failures and resume from checkpoints instead of restarting from scratch in this 22-minute conference talk. Discover why stateless architectures that work for simple demos become problematic at scale, causing loss of compute, progress, and user trust when long-running workflows fail. Explore how companies like OpenAI use Temporal for products like Deep Research to create durable agents that recover gracefully from failures. Master building resilient AI agents with PydanticAI and Temporal, implementing durable execution patterns, and gaining production-grade observability through Pydantic Logfire and Evals. Understand how to compose multi-agent systems that handle failures gracefully and stop wasting resources on failed agent runs that restart from the beginning. Follow along with real code examples, including a Deep Research implementation that demonstrates transforming fragile prototypes into robust production systems through proper architectural patterns.