Durable Agents: long running AI workflows in a flakey world

As AI gets more powerful the limit on the magnitude of tasks it can perform continues to increase.

But that leads to a problem: what happens when something goes wrong? The time taken to recover and the value of the lost work is increasing too.

Most AI agents today are trapped in the "chat paradigm"—they're synchronous, short-lived, and frustratingly limited. In this session, we'll talk about how durable execution makes it possible to break free from these constraints, using Pydantic AI's Temporal integration to concretely demonstrate what this looks like in practice. You'll learn concrete patterns for state management, streaming results, and failure recovery that transform agents from impressive toys into production-grade systems that deliver real economic value.