Prefect provides a novel approach to durable execution systems by decoupling durability from workflow identity, allowing results to be addressed by cache keys rather than the workflow that produced them. This enables efficient recovery from failures and coordination of concurrent executions, especially in data systems where the unit of work may not be tied to a specific workflow. Prefect's design allows results to be stored in user-controlled object storage systems like S3 or Azure Blob, making them accessible and reusable across different workflows and contexts. With its support for cache policies, lock managers, and transactional semantics with rollback hooks, Prefect facilitates exactly-once execution across workflows and ensures atomic operations can succeed or fail as a group. Additionally, Prefect separates code from compute by allowing flows to be submitted to dynamically provisioned infrastructure, adapting resource allocation based on runtime needs rather than deployment constraints. This composability and flexibility make Prefect particularly suitable for data systems that require big computations and expensive resources, offering a range of primitives that can be tailored to specific durability semantics.