LLMs are inherently nondeterministic: give a model the same prompt twice, and you may get two different answers. Temporal Workflows, by contrast, are deterministic by design. At first, those ideas may seem incompatible. In practice, they’re complementary: Thousands of organizations rely on Temporal's Durable Execution to build reliable AI applications on top of inherently unpredictable models.
How can deterministic Workflow code accommodate a model that behaves differently each time you call it? The answer is History Replay, the mechanism Temporal uses to recover the state of a running application after a crash or restart. In this session, we'll break down how replay works, how Event History records what happened, and why deterministic Workflow code makes reliable recovery possible. We'll also look at how Temporal safely handles nondeterministic operations such as LLM calls, API requests, randomness, time, and human input without losing state or progress.
By the end of this session, you'll have a practical mental model for determinism, replay, and Durable Execution, along with a clear understanding of how Temporal provides a reliable foundation for your own applications.
Can't make the live session? Register anyway, and we'll send the recording to your inbox.
Principal Developer Advocate
Temporal Technologies
Alternating between software engineering and technical education roles, Tom Wheeler's career spans nearly 30 years in the financial, healthcare, aerospace, and tech industries.
Prior to joining Temporal in 2022, Tom wrote and taught training courses at Cloudera, developed engineering software at Boeing, designed and implemented a distributed data processing system at WebMD, and created some of the earliest web applications at brokerage firm A.G. Edwards.
Tom enjoys learning new things and teaching them to others, which has led him to deliver presentations and workshops at conferences around the world, including Replay, We Are Developers, Great International Developer Summit, APIDays Paris, Strata, OSCON, JavaOne, and Strange Loop.
