This talk explores how Cursor is building cloud-based AI agents powered by Temporal to enable scalable, persistent workflows. It walks through the evolution from local multi-agent experimentation to a cloud architecture where each agent runs in its own VM and is orchestrated via Temporal workflows. The session also covers key infrastructure challenges—like long-running workflows, versioning, and deployment stability—and how the team redesigned their system to improve reliability and observability. Finally, it highlights real-world use cases, internal adoption, and future directions for extending Temporal-backed agents.